diff --git a/.github/workflows/sync-models.yml b/.github/workflows/sync-models.yml
index ac9f3690bcd..3382485eb62 100644
--- a/.github/workflows/sync-models.yml
+++ b/.github/workflows/sync-models.yml
@@ -74,6 +74,7 @@ jobs:
           GH_TOKEN: ${{ github.token }}
           ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
           BASETEN_API_KEY: ${{ secrets.BASETEN_API_KEY }}
+          CHEAPERINFERENCE_API_KEY: ${{ secrets.CHEAPERINFERENCE_API_KEY }}
           DEEPINFRA_API_KEY: ${{ secrets.DEEPINFRA_API_KEY }}
           DIGITALOCEAN_API_TOKEN: ${{ secrets.DIGITALOCEAN_API_TOKEN }}
           DIGITALOCEAN_ACCESS_TOKEN: ${{ secrets.DIGITALOCEAN_ACCESS_TOKEN }}
diff --git a/packages/core/src/sync/index.ts b/packages/core/src/sync/index.ts
index 7f88aacb5bb..584b243f9fe 100644
--- a/packages/core/src/sync/index.ts
+++ b/packages/core/src/sync/index.ts
@@ -9,6 +9,7 @@ import { MissingReasoningOptionsError } from "./missing-reasoning-options.js";
 import { ambient } from "./providers/ambient.js";
 import { anthropic } from "./providers/anthropic.js";
 import { baseten } from "./providers/baseten.js";
+import { cheaperinference } from "./providers/cheaperinference.js";
 import { chutes } from "./providers/chutes.js";
 import { cloudflareAiGateway } from "./providers/cloudflare-ai-gateway.js";
 import { cloudflareWorkersAi } from "./providers/cloudflare-workers-ai.js";
@@ -133,6 +134,7 @@ export const providers: {
   ambient: SyncProvider<any>;
   anthropic: SyncProvider<any>;
   baseten: SyncProvider<any>;
+  cheaperinference: SyncProvider<any>;
   chutes: SyncProvider<any>;
   "cloudflare-ai-gateway": SyncProvider<any>;
   "cloudflare-workers-ai": SyncProvider<any>;
@@ -168,6 +170,7 @@ export const providers: {
   ambient,
   anthropic,
   baseten,
+  cheaperinference,
   chutes,
   "cloudflare-ai-gateway": cloudflareAiGateway,
   "cloudflare-workers-ai": cloudflareWorkersAi,
@@ -203,6 +206,7 @@ export const providers: {
 
 export const groups = {
   aggregators: [
+    "cheaperinference",
     "crossmodel",
     "edenai",
     "empiriolabs",
diff --git a/packages/core/src/sync/providers/cheaperinference.ts b/packages/core/src/sync/providers/cheaperinference.ts
new file mode 100644
index 00000000000..4ede4f2d823
--- /dev/null
+++ b/packages/core/src/sync/providers/cheaperinference.ts
@@ -0,0 +1,195 @@
+import { z } from "zod";
+
+import type { ExistingModel, SyncedFullModel, SyncedModel, SyncProvider } from "../index.js";
+import { MissingReasoningOptionsError } from "../missing-reasoning-options.js";
+import { factorBaseModel } from "./openrouter.js";
+
+const API_ENDPOINT = "https://api.cheaperinference.com/v1/models";
+
+/** Long-context band. The gateway bills prompts above the threshold at these rates. */
+const AboveThreshold = z
+  .object({
+    input_token_price_threshold: z.number().int().positive(),
+    input_per_million: z.string().min(1),
+    output_per_million: z.string().min(1),
+    cache_read_input_per_million: z.string().min(1).nullish(),
+    cache_write_input_per_million: z.string().min(1).nullish(),
+  })
+  .passthrough();
+
+const CheaperInferencePricing = z
+  .object({
+    currency: z.literal("USD"),
+    input_per_million: z.string().min(1),
+    output_per_million: z.string().min(1),
+    cache_read_input_per_million: z.string().min(1).nullish(),
+    cache_write_input_per_million: z.string().min(1).nullish(),
+    input_token_price_threshold: z.number().int().positive().nullish(),
+    above_threshold: AboveThreshold.nullish(),
+  })
+  .passthrough();
+
+export const CheaperInferenceModel = z
+  .object({
+    id: z.string().min(1),
+    object: z.literal("model"),
+    type: z.string().min(1),
+    endpoint: z.string().min(1),
+    context_length: z.number().int().positive().nullish(),
+    max_output_tokens: z.number().int().positive().nullish(),
+    is_free: z.boolean(),
+    pricing: CheaperInferencePricing,
+  })
+  .passthrough();
+
+export const CheaperInferenceResponse = z
+  .object({
+    object: z.literal("list"),
+    data: z.array(CheaperInferenceModel),
+    pricing_version: z.string().min(1),
+    pricing_checked_at: z.string().min(1),
+  })
+  .passthrough();
+
+export type CheaperInferenceModel = z.infer<typeof CheaperInferenceModel>;
+
+export const cheaperinference = {
+  id: "cheaperinference",
+  name: "CheaperInference",
+  modelsDir: "providers/cheaperinference/models",
+  // The catalog carries pricing and limits but no reasoning controls, so new
+  // models need hand-authored reasoning_options before they can be created.
+  skipCreates: true,
+  sourceID(model) {
+    return isTokenPricedTextModel(model) ? model.id : undefined;
+  },
+  skippedNotice(ids) {
+    if (ids.length === 0) return [];
+    return [
+      `${ids.length} CheaperInference models were not created because the catalog exposes no reasoning controls, which this repo requires for reasoning models.`,
+      `Skipped remote IDs: ${ids.map((id) => `\`${id}\``).join(", ")}`,
+    ];
+  },
+  async fetchModels() {
+    return fetchCheaperInferenceModels();
+  },
+  parseModels(raw) {
+    return CheaperInferenceResponse.parse(raw).data;
+  },
+  translateModel(model, context) {
+    const existing = context.existing(model.id);
+    if (existing === undefined) return undefined;
+    return {
+      id: model.id,
+      model: buildCheaperInferenceModel(model, existing),
+    };
+  },
+} satisfies SyncProvider<CheaperInferenceModel>;
+
+export async function fetchCheaperInferenceModels(fetcher: typeof fetch = fetch) {
+  const apiKey = process.env["CHEAPERINFERENCE_API_KEY"];
+  if (apiKey === undefined || apiKey === "") {
+    throw new Error("CHEAPERINFERENCE_API_KEY is required to read the CheaperInference catalog");
+  }
+  const response = await fetcher(API_ENDPOINT, {
+    headers: { Authorization: `Bearer ${apiKey}` },
+  });
+  if (!response.ok) {
+    throw new Error(
+      `CheaperInference models request failed: ${response.status} ${response.statusText}`,
+    );
+  }
+  return CheaperInferenceResponse.parse(await response.json());
+}
+
+/**
+ * Only text models billed per token can be expressed by the catalog schema.
+ * Image and video routes price per unit of generated media, and free routes
+ * carry no rates worth syncing.
+ */
+function isTokenPricedTextModel(model: CheaperInferenceModel) {
+  return (
+    model.type === "text" &&
+    model.endpoint === "/v1/chat/completions" &&
+    !model.is_free &&
+    price(model.pricing.input_per_million) > 0 &&
+    price(model.pricing.output_per_million) > 0
+  );
+}
+
+function price(value: string) {
+  const parsed = Number(value);
+  if (!Number.isFinite(parsed) || parsed < 0) {
+    throw new Error(`CheaperInference returned an unusable price: ${value}`);
+  }
+  return parsed;
+}
+
+function optionalPrice(value: string | null | undefined) {
+  return value === null || value === undefined ? undefined : price(value);
+}
+
+export function buildCheaperInferenceModel(
+  model: CheaperInferenceModel,
+  existing: ExistingModel,
+): SyncedModel {
+  if (existing.reasoning !== false && existing.reasoning_options === undefined) {
+    // The runner keeps the local file and reports the ID instead of failing the
+    // whole sync, so one unresearched model cannot stop the hourly run.
+    throw new MissingReasoningOptionsError(
+      model.id,
+      "reasons on this host but the catalog exposes no reasoning controls, so reasoning_options must be hand-authored",
+    );
+  }
+
+  const { base_model: baseModel, base_model_omit: baseModelOmit, ...current } = existing;
+  const pricing = model.pricing;
+
+  const cost = {
+    ...existing.cost,
+    input: price(pricing.input_per_million),
+    output: price(pricing.output_per_million),
+    cache_read: optionalPrice(pricing.cache_read_input_per_million),
+    cache_write: optionalPrice(pricing.cache_write_input_per_million),
+    tiers: buildTiers(pricing.above_threshold),
+  };
+
+  // The gateway is authoritative for what it serves, so its limits win when it
+  // publishes them. Missing values stay inherited from the lab entry.
+  const limit = {
+    ...existing.limit,
+    ...(model.context_length === null || model.context_length === undefined
+      ? {}
+      : { context: model.context_length }),
+    ...(model.max_output_tokens === null || model.max_output_tokens === undefined
+      ? {}
+      : { output: model.max_output_tokens }),
+  };
+
+  const values = {
+    ...current,
+    cost,
+    limit,
+  } as SyncedFullModel;
+
+  return baseModel === undefined
+    ? values
+    : factorBaseModel(baseModel, values, limit, baseModelOmit);
+}
+
+function buildTiers(above: z.infer<typeof AboveThreshold> | null | undefined) {
+  if (above === null || above === undefined) return undefined;
+  // The threshold is the last token billed at the base rate, so the band starts
+  // one token later. The gateway reports both 271_999 and 272_000 for the same
+  // 272k boundary, which normalises to one tier size.
+  const size = above.input_token_price_threshold + 1;
+  return [
+    {
+      tier: { type: "context" as const, size: size - (size % 1000) },
+      input: price(above.input_per_million),
+      output: price(above.output_per_million),
+      cache_read: optionalPrice(above.cache_read_input_per_million),
+      cache_write: optionalPrice(above.cache_write_input_per_million),
+    },
+  ];
+}
diff --git a/packages/core/test/cheaperinference.test.ts b/packages/core/test/cheaperinference.test.ts
new file mode 100644
index 00000000000..f2265ea9e32
--- /dev/null
+++ b/packages/core/test/cheaperinference.test.ts
@@ -0,0 +1,179 @@
+import { expect, spyOn, test } from "bun:test";
+import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
+import path from "node:path";
+
+import { syncProvider } from "../src/sync/index.js";
+import * as missingIssues from "../src/sync/missing-issues.js";
+import { MissingReasoningOptionsError } from "../src/sync/missing-reasoning-options.js";
+import {
+  buildCheaperInferenceModel,
+  CheaperInferenceResponse,
+  cheaperinference,
+} from "../src/sync/providers/cheaperinference.js";
+
+function sourceModel(overrides: Record<string, unknown> = {}) {
+  return {
+    id: "claude-sonnet-5",
+    object: "model" as const,
+    type: "text",
+    endpoint: "/v1/chat/completions",
+    context_length: 1_000_000,
+    max_output_tokens: 64_000,
+    is_free: false,
+    pricing: {
+      currency: "USD" as const,
+      input_per_million: "1.400000",
+      output_per_million: "7.000000",
+      cache_read_input_per_million: "0.140000",
+      cache_write_input_per_million: "1.700000",
+      input_token_price_threshold: null,
+      above_threshold: null,
+    },
+    ...overrides,
+  };
+}
+
+const existing = {
+  base_model: "anthropic/claude-sonnet-5",
+  reasoning: true,
+  reasoning_options: [{ type: "effort" as const, values: ["low", "high"] }],
+  cost: { input: 1.4, output: 7 },
+  limit: { context: 1_000_000, output: 64_000 },
+};
+
+test("syncs the gateway's discounted rates, including cache rates", () => {
+  const model = buildCheaperInferenceModel(sourceModel() as never, existing as never);
+
+  expect(model).toMatchObject({
+    base_model: "anthropic/claude-sonnet-5",
+    cost: { input: 1.4, output: 7, cache_read: 0.14, cache_write: 1.7 },
+  });
+});
+
+test("turns a long-context band into a single context tier", () => {
+  const model = buildCheaperInferenceModel(
+    sourceModel({
+      id: "gpt-5.6-luna",
+      context_length: 1_050_000,
+      max_output_tokens: 128_000,
+      pricing: {
+        currency: "USD",
+        input_per_million: "0.080000",
+        output_per_million: "0.480000",
+        cache_read_input_per_million: "0.008000",
+        cache_write_input_per_million: "0.100000",
+        input_token_price_threshold: 271_999,
+        above_threshold: {
+          input_token_price_threshold: 271_999,
+          input_per_million: "0.160000",
+          output_per_million: "0.720000",
+          cache_read_input_per_million: "0.016000",
+          cache_write_input_per_million: "0.200000",
+        },
+      },
+    }) as never,
+    { ...existing, base_model: "openai/gpt-5.6-luna" } as never,
+  );
+
+  expect(model.cost?.tiers).toEqual([
+    {
+      tier: { type: "context", size: 272_000 },
+      input: 0.16,
+      output: 0.72,
+      cache_read: 0.016,
+      cache_write: 0.2,
+    },
+  ]);
+});
+
+test("takes the gateway's limits when it publishes them and inherits when it does not", () => {
+  const capped = buildCheaperInferenceModel(
+    sourceModel({ id: "deepseek-v4-flash-0731", context_length: 1_048_576, max_output_tokens: 65_536 }) as never,
+    { ...existing, base_model: "deepseek/deepseek-v4-flash-0731", limit: { context: 1_000_000, output: 384_000 } } as never,
+  );
+  expect(capped.limit).toMatchObject({ context: 1_048_576, output: 65_536 });
+
+  // With nothing published, the synced file authors no limit override at all
+  // and the lab entry's limits keep applying.
+  const unpublished = buildCheaperInferenceModel(
+    sourceModel({ id: "gpt-oss-120b", context_length: null, max_output_tokens: null }) as never,
+    { ...existing, base_model: "openai/gpt-oss-120b", limit: { context: 131_072, output: 32_768 } } as never,
+  );
+  expect(unpublished.limit).toBeUndefined();
+});
+
+test("refuses to sync a reasoning model that has no authored controls", () => {
+  // MissingReasoningOptionsError, not a bare Error: the runner catches this one
+  // to skip the single ID and keep the local file, instead of failing the run.
+  let thrown: unknown;
+  try {
+    buildCheaperInferenceModel(sourceModel() as never, {
+      ...existing,
+      reasoning_options: undefined,
+    } as never);
+  } catch (error) {
+    thrown = error;
+  }
+  expect(thrown).toBeInstanceOf(MissingReasoningOptionsError);
+  expect((thrown as MissingReasoningOptionsError).modelId).toBe("claude-sonnet-5");
+  expect((thrown as Error).message).toContain("reasoning_options");
+});
+
+test("an unresearched reasoner is reported and its file kept, and the rest of the run continues", async () => {
+  const dir = await mkdtemp(path.join(import.meta.dirname, "../../../providers/.cheaperinference-sync-"));
+  const modelsDir = path.join(dir, "models");
+  const authored = '# authored by hand\nbase_model = "anthropic/claude-sonnet-5"\n';
+  const researched =
+    '# authored by hand\nbase_model = "anthropic/claude-opus-5"\n\n[[reasoning_options]]\ntype = "effort"\nvalues = ["low", "high"]\n';
+  await mkdir(modelsDir, { recursive: true });
+  await writeFile(path.join(modelsDir, "claude-sonnet-5.toml"), authored);
+  await writeFile(path.join(modelsDir, "claude-opus-5.toml"), researched);
+  const issues = spyOn(missingIssues, "openMissingModelIssues").mockResolvedValue([]);
+  try {
+    const result = await syncProvider(
+      {
+        ...cheaperinference,
+        modelsDir,
+        async fetchModels() {
+          return {
+            object: "list",
+            data: [sourceModel(), sourceModel({ id: "claude-opus-5" })],
+            pricing_version: "sha256:test",
+            pricing_checked_at: "2026-09-16T08:00:58.592Z",
+          } as never;
+        },
+      },
+      { openIssues: true },
+    );
+    expect(result.deleted).toBe(0);
+    expect(await readFile(path.join(modelsDir, "claude-sonnet-5.toml"), "utf8")).toBe(authored);
+    expect(issues.mock.calls[0]?.[1]).toEqual(["claude-sonnet-5"]);
+    expect(issues.mock.calls[0]?.[2]?.reasons?.["claude-sonnet-5"]).toContain("reasoning_options");
+  } finally {
+    issues.mockRestore();
+    await rm(dir, { recursive: true, force: true });
+  }
+});
+
+test("only syncs token-priced text routes", () => {
+  const ids = [
+    sourceModel(),
+    sourceModel({ id: "nano-banana-2", type: "image", endpoint: "/v1/images/generations" }),
+    sourceModel({ id: "seedance-2.0", type: "video", endpoint: "/v1/videos/generations" }),
+    sourceModel({ id: "some-free-model", is_free: true }),
+  ].map((model) => cheaperinference.sourceID(model as never));
+
+  expect(ids).toEqual(["claude-sonnet-5", undefined, undefined, undefined]);
+});
+
+test("parses the catalog response shape", () => {
+  const parsed = CheaperInferenceResponse.parse({
+    object: "list",
+    data: [sourceModel()],
+    pricing_version: "sha256:abc",
+    pricing_checked_at: "2026-09-10T21:01:09.012Z",
+    pricing_updated_at: "2026-08-05T00:00:00.000Z",
+  });
+
+  expect(parsed.data[0]?.id).toBe("claude-sonnet-5");
+});
diff --git a/providers/cheaperinference/logo.svg b/providers/cheaperinference/logo.svg
new file mode 100644
index 00000000000..36847516c91
--- /dev/null
+++ b/providers/cheaperinference/logo.svg
@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="currentColor">
+  <path d="M9.107 2.786H15.429V8.839H9.589a3.213 3.213 0 0 0 0.321 6.375H15.214L9.434 20.994A9.107 9.107 0 1 1 9.107 2.786Z"/>
+  <path d="M22.179 3.429H22.607a1.286 1.286 0 0 1 1.286 1.286V19.714a1.393 1.393 0 0 1 -1.393 1.393H17.464V8.143Z"/>
+</svg>
diff --git a/providers/cheaperinference/models/claude-fable-5.1.toml b/providers/cheaperinference/models/claude-fable-5.1.toml
new file mode 100644
index 00000000000..a28cc5472b2
--- /dev/null
+++ b/providers/cheaperinference/models/claude-fable-5.1.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "anthropic/claude-fable-5-1"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.175
+cache_write = 8.5
diff --git a/providers/cheaperinference/models/claude-fable-5.toml b/providers/cheaperinference/models/claude-fable-5.toml
new file mode 100644
index 00000000000..fb04e3bf7be
--- /dev/null
+++ b/providers/cheaperinference/models/claude-fable-5.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "anthropic/claude-fable-5"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.7
+cache_write = 8.5
diff --git a/providers/cheaperinference/models/claude-haiku-4.5.toml b/providers/cheaperinference/models/claude-haiku-4.5.toml
new file mode 100644
index 00000000000..8066af50629
--- /dev/null
+++ b/providers/cheaperinference/models/claude-haiku-4.5.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: 2 of 4 colour probes exact, the others "lime" for green and
+# "magenta" for purple (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.0/$5.0 per 1M).
+base_model = "anthropic/claude-haiku-4-5"
+reasoning = false
+
+[cost]
+input = 0.7
+output = 3.5
+cache_read = 0.07
+cache_write = 0.85
diff --git a/providers/cheaperinference/models/claude-opus-4-8-fast.toml b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
new file mode 100644
index 00000000000..6ce5f09b443
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4-8-fast.toml
@@ -0,0 +1,26 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 400 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($12.0/$60.0 per 1M).
+base_model = "anthropic/claude-opus-4-8"
+attachment = false
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 8.4
+output = 42.0
+cache_read = 0.84
+cache_write = 10.2
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/claude-opus-4.5.toml b/providers/cheaperinference/models/claude-opus-4.5.toml
new file mode 100644
index 00000000000..4586bb922b2
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.5.toml
@@ -0,0 +1,19 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-5"
+reasoning = false
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
+
+[limit]
+context = 198000
+output = 32768
diff --git a/providers/cheaperinference/models/claude-opus-4.6.toml b/providers/cheaperinference/models/claude-opus-4.6.toml
new file mode 100644
index 00000000000..07f499ad23d
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.6.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-6"
+reasoning = false
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-opus-4.7.toml b/providers/cheaperinference/models/claude-opus-4.7.toml
new file mode 100644
index 00000000000..4af692312f0
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.7.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-7"
+reasoning = false
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-opus-4.8.toml b/providers/cheaperinference/models/claude-opus-4.8.toml
new file mode 100644
index 00000000000..e1597404760
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-4.8.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-4-8"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-opus-5-fast.toml b/providers/cheaperinference/models/claude-opus-5-fast.toml
new file mode 100644
index 00000000000..df7310cc5cd
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-5-fast.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "anthropic/claude-opus-5"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.7
+cache_write = 8.5
diff --git a/providers/cheaperinference/models/claude-opus-5.toml b/providers/cheaperinference/models/claude-opus-5.toml
new file mode 100644
index 00000000000..3ae53009f66
--- /dev/null
+++ b/providers/cheaperinference/models/claude-opus-5.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$25.0 per 1M).
+base_model = "anthropic/claude-opus-5"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 3.5
+output = 17.5
+cache_read = 0.35
+cache_write = 4.25
diff --git a/providers/cheaperinference/models/claude-sonnet-4.5.toml b/providers/cheaperinference/models/claude-sonnet-4.5.toml
new file mode 100644
index 00000000000..28babdc5771
--- /dev/null
+++ b/providers/cheaperinference/models/claude-sonnet-4.5.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: 3 of 4 colour probes exact and the fourth called "lime" for
+# green (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
+base_model = "anthropic/claude-sonnet-4-5"
+reasoning = false
+
+[cost]
+input = 2.1
+output = 10.5
+cache_read = 0.21
+cache_write = 2.55
+
+[limit]
+context = 198000
diff --git a/providers/cheaperinference/models/claude-sonnet-4.6.toml b/providers/cheaperinference/models/claude-sonnet-4.6.toml
new file mode 100644
index 00000000000..d04b32f738a
--- /dev/null
+++ b/providers/cheaperinference/models/claude-sonnet-4.6.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort and thinking.budget_tokens each return
+# 200 with no reasoning_content and no reasoning token count, though the catalog reports
+# capabilities.reasoning true (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
+base_model = "anthropic/claude-sonnet-4-6"
+reasoning = false
+
+[cost]
+input = 2.1
+output = 10.5
+cache_read = 0.21
+cache_write = 2.55
diff --git a/providers/cheaperinference/models/claude-sonnet-5.toml b/providers/cheaperinference/models/claude-sonnet-5.toml
new file mode 100644
index 00000000000..e55c244600a
--- /dev/null
+++ b/providers/cheaperinference/models/claude-sonnet-5.toml
@@ -0,0 +1,25 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$10.0 per 1M).
+base_model = "anthropic/claude-sonnet-5"
+attachment = false
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 1.4
+output = 7.0
+cache_read = 0.14
+cache_write = 1.7
+
+[limit]
+output = 64000
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash-0731.toml b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
new file mode 100644
index 00000000000..a9301d2e38a
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-flash-0731.toml
@@ -0,0 +1,31 @@
+# Image input works on this host although the base model is text-only: a 32x32 solid-colour PNG as a
+# data: URI was named correctly on 10 of 10 probes across four colours, so attachment and the input
+# modalities are overridden here rather than inherited (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 40.00% off the model maker's list price ($0.076/$0.153 per 1M).
+base_model = "deepseek/deepseek-v4-flash-0731"
+attachment = true
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.0456
+output = 0.0918
+cache_read = 0.00646
+cache_write = 0.0456
+
+[limit]
+context = 1048576
+output = 65536
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/cheaperinference/models/deepseek-v4-flash.toml b/providers/cheaperinference/models/deepseek-v4-flash.toml
new file mode 100644
index 00000000000..2fd837482f3
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-flash.toml
@@ -0,0 +1,25 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# NOIMAGE for green and purple on 4 of 4 probes, with only the blue guess landing (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.09% off the model maker's list price ($0.13/$0.26 per 1M).
+base_model = "deepseek/deepseek-v4-flash"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high", "xhigh"]
+
+[cost]
+input = 0.071388
+output = 0.142777
+cache_read = 0.01105
+cache_write = 0.071388
+
+[limit]
+output = 128000
diff --git a/providers/cheaperinference/models/deepseek-v4-pro-0813.toml b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
new file mode 100644
index 00000000000..bba5653d0a2
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-pro-0813.toml
@@ -0,0 +1,25 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# NOIMAGE for green and purple on 4 of 4 probes, with only the blue guess landing (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
+base_model = "deepseek/deepseek-v4-pro-0813"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high", "max"]
+
+[cost]
+input = 0.462
+output = 1.386
+cache_read = 0.01541
+cache_write = 0.462
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/deepseek-v4-pro.toml b/providers/cheaperinference/models/deepseek-v4-pro.toml
new file mode 100644
index 00000000000..6c999fa5252
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4-pro.toml
@@ -0,0 +1,25 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 6 of 6 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.66/$1.98 per 1M).
+base_model = "deepseek/deepseek-v4-pro"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high", "xhigh"]
+
+[cost]
+input = 0.462
+output = 1.386
+cache_read = 0.038981
+cache_write = 0.462
+
+[limit]
+output = 128000
diff --git a/providers/cheaperinference/models/deepseek-v4.1-flash.toml b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
new file mode 100644
index 00000000000..0483ac30e78
--- /dev/null
+++ b/providers/cheaperinference/models/deepseek-v4.1-flash.toml
@@ -0,0 +1,25 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
+base_model = "deepseek/deepseek-v4.1-flash"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.21
+output = 0.84
+cache_read = 0.0042
+cache_write = 0.21
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/gemini-2.5-flash.toml b/providers/cheaperinference/models/gemini-2.5-flash.toml
new file mode 100644
index 00000000000..82054bb36ca
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-2.5-flash.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
+base_model = "google/gemini-2.5-flash"
+reasoning = false
+
+[cost]
+input = 0.21
+output = 1.75
+cache_read = 0.021
+cache_write = 0.05831
+
+[limit]
+output = 65535
diff --git a/providers/cheaperinference/models/gemini-3-5-flash.toml b/providers/cheaperinference/models/gemini-3-5-flash.toml
new file mode 100644
index 00000000000..974352eeb29
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3-5-flash.toml
@@ -0,0 +1,21 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.5/$9.0 per 1M).
+base_model = "google/gemini-3.5-flash"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 1.05
+output = 6.3
+cache_read = 0.105
+cache_write = 0.05831
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/gemini-3-flash-preview.toml b/providers/cheaperinference/models/gemini-3-flash-preview.toml
new file mode 100644
index 00000000000..d892bebf1c7
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3-flash-preview.toml
@@ -0,0 +1,21 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.5/$3.0 per 1M).
+base_model = "google/gemini-3-flash-preview"
+
+[interleaved]
+field = "reasoning_details"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.35
+output = 2.1
+cache_read = 0.035
+cache_write = 0.05831
+
+[limit]
+context = 256000
diff --git a/providers/cheaperinference/models/gemini-3.1-flash-lite.toml b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
new file mode 100644
index 00000000000..8600ee38911
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.1-flash-lite.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# No reasoning on this host: plain, reasoning.enabled, reasoning.effort, top-level reasoning_effort and
+# thinking.budget_tokens each return 200 with no reasoning_content and no reasoning token count; the catalog
+# reports capabilities.reasoning false (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$1.5 per 1M).
+base_model = "google/gemini-3.1-flash-lite"
+reasoning = false
+
+[cost]
+input = 0.201421
+output = 1.20853
+cache_read = 0.02125
+cache_write = 0.067113
diff --git a/providers/cheaperinference/models/gemini-3.1-pro-preview.toml b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
new file mode 100644
index 00000000000..f312c3a785d
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.1-pro-preview.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$12.0 per 1M).
+base_model = "google/gemini-3.1-pro-preview"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 1.4
+output = 8.4
+cache_read = 0.14
+cache_write = 0.2625
+
+[limit]
+context = 1000000
+output = 32768
diff --git a/providers/cheaperinference/models/gemini-3.6-flash.toml b/providers/cheaperinference/models/gemini-3.6-flash.toml
new file mode 100644
index 00000000000..f4eee3e05ea
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.6-flash.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Reasoning is billed and counted but never returned: 8 of 8 probes reported
+# usage.completion_tokens_details.reasoning_tokens (235-385) while the message carried only content,
+# so there is no side channel to declare here (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 15.00% off the model maker's list price ($0.75/$3.75 per 1M).
+base_model = "google/gemini-3.6-flash"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.6375
+output = 3.1875
+cache_read = 0.06375
+cache_write = 0.035416
diff --git a/providers/cheaperinference/models/gemini-3.7-flash.toml b/providers/cheaperinference/models/gemini-3.7-flash.toml
new file mode 100644
index 00000000000..10fbb304fff
--- /dev/null
+++ b/providers/cheaperinference/models/gemini-3.7-flash.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$3.75 per 1M).
+base_model = "google/gemini-3.7-flash"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 0.525
+output = 2.625
+cache_read = 0.0525
+cache_write = 0.02919
diff --git a/providers/cheaperinference/models/glm-4.5-air.toml b/providers/cheaperinference/models/glm-4.5-air.toml
new file mode 100644
index 00000000000..a94d1e11ce8
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.5-air.toml
@@ -0,0 +1,22 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: thinking.type = enabled|disabled
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.2/$1.1 per 1M).
+base_model = "zhipuai/glm-4.5-air"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.104739
+output = 0.605
+cache_read = 0.017
+cache_write = 0.104739
+
+[limit]
+context = 128000
+output = 32768
diff --git a/providers/cheaperinference/models/glm-4.5.toml b/providers/cheaperinference/models/glm-4.5.toml
new file mode 100644
index 00000000000..92935999ec3
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.5.toml
@@ -0,0 +1,22 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
+base_model = "zhipuai/glm-4.5"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.33
+output = 1.21
+cache_read = 0.051
+cache_write = 0.33
+
+[limit]
+context = 128000
+output = 32768
diff --git a/providers/cheaperinference/models/glm-4.6.toml b/providers/cheaperinference/models/glm-4.6.toml
new file mode 100644
index 00000000000..6d000fa4889
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.6.toml
@@ -0,0 +1,22 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
+base_model = "zhipuai/glm-4.6"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.33
+output = 1.21
+cache_read = 0.051
+cache_write = 0.33
+
+[limit]
+context = 198000
+output = 16384
diff --git a/providers/cheaperinference/models/glm-4.7.toml b/providers/cheaperinference/models/glm-4.7.toml
new file mode 100644
index 00000000000..aa25e6b9ae7
--- /dev/null
+++ b/providers/cheaperinference/models/glm-4.7.toml
@@ -0,0 +1,20 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.6/$2.2 per 1M).
+base_model = "zhipuai/glm-4.7"
+interleaved = true
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.322274
+output = 1.21
+cache_read = 0.051
+cache_write = 0.322274
+
+[limit]
+context = 128000
+output = 32768
diff --git a/providers/cheaperinference/models/glm-5.1.toml b/providers/cheaperinference/models/glm-5.1.toml
new file mode 100644
index 00000000000..2de36652906
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.1.toml
@@ -0,0 +1,21 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 4 of 4 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
+base_model = "zhipuai/glm-5.1"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.77
+output = 2.42
+cache_read = 0.119
+cache_write = 0.77
+
+[limit]
+output = 24000
diff --git a/providers/cheaperinference/models/glm-5.2.toml b/providers/cheaperinference/models/glm-5.2.toml
new file mode 100644
index 00000000000..18aa327e9ed
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.2.toml
@@ -0,0 +1,21 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($0.8/$2.55 per 1M).
+base_model = "zhipuai/glm-5.2"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["high", "max"]
+
+[cost]
+input = 0.44
+output = 1.4025
+cache_read = 0.068
+cache_write = 0.44
+
+[limit]
+context = 1048576
diff --git a/providers/cheaperinference/models/glm-5.3-flash.toml b/providers/cheaperinference/models/glm-5.3-flash.toml
new file mode 100644
index 00000000000..2b0a819a45b
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.3-flash.toml
@@ -0,0 +1,25 @@
+# No dependable image input on this host: 10 probes gave 3 colours right, "red" for green and one
+# NOIMAGE, so image input cannot be relied on here (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 51.66% off the model maker's list price ($0.15/$0.5 per 1M).
+base_model = "zhipuai/glm-5.3-flash"
+attachment = false
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.072511
+output = 0.241706
+cache_read = 0.01275
+cache_write = 0.072511
+
+[limit]
+context = 1048576
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/glm-5.3.toml b/providers/cheaperinference/models/glm-5.3.toml
new file mode 100644
index 00000000000..22332be6ffc
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.3.toml
@@ -0,0 +1,18 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 45.00% off the model maker's list price ($1.4/$4.4 per 1M).
+base_model = "zhipuai/glm-5.3"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 0.77
+output = 2.42
+cache_read = 0.119
+cache_write = 0.77
diff --git a/providers/cheaperinference/models/glm-5.toml b/providers/cheaperinference/models/glm-5.toml
new file mode 100644
index 00000000000..d013708de60
--- /dev/null
+++ b/providers/cheaperinference/models/glm-5.toml
@@ -0,0 +1,22 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 502 on 6 of 6 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 51.66% off the model maker's list price ($1.0/$3.2 per 1M).
+base_model = "zhipuai/glm-5"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.483412
+output = 1.546919
+cache_read = 0.085
+cache_write = 0.483412
+
+[limit]
+context = 198000
+output = 32768
diff --git a/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml b/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
new file mode 100644
index 00000000000..8c0e1534430
--- /dev/null
+++ b/providers/cheaperinference/models/google/gemini-3.5-flash-lite.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$2.5 per 1M).
+base_model = "google/gemini-3.5-flash-lite"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.21
+output = 1.75
+cache_read = 0.021
+cache_write = 0.078333
diff --git a/providers/cheaperinference/models/gpt-4.1-nano.toml b/providers/cheaperinference/models/gpt-4.1-nano.toml
new file mode 100644
index 00000000000..eea038d8246
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-4.1-nano.toml
@@ -0,0 +1,15 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 15.00% off the model maker's list price ($0.1/$0.4 per 1M).
+base_model = "openai/gpt-4.1-nano"
+attachment = false
+
+[cost]
+input = 0.085
+output = 0.34
+cache_read = 0.008499
+cache_write = 0.085
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5-mini.toml b/providers/cheaperinference/models/gpt-5-mini.toml
new file mode 100644
index 00000000000..579cbddba16
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5-mini.toml
@@ -0,0 +1,19 @@
+# No usable image input on this host: the request is accepted but the colour is guessed - 0 of 6 probes
+# right (orange, maroon, olive, magenta, cyan, turquoise) (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.25/$2.0 per 1M).
+base_model = "openai/gpt-5-mini"
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.201421
+output = 1.611374
+cache_read = 0.020142
+cache_write = 0.201421
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5-nano.toml b/providers/cheaperinference/models/gpt-5-nano.toml
new file mode 100644
index 00000000000..dc2627003cb
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5-nano.toml
@@ -0,0 +1,19 @@
+# No usable image input on this host: the request is accepted but the colour is guessed - 1 of 6 probes
+# right, the rest wrong or empty (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.05/$0.4 per 1M).
+base_model = "openai/gpt-5-nano"
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high"]
+
+[cost]
+input = 0.040284
+output = 0.322274
+cache_read = 0.004028
+cache_write = 0.040284
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.2-codex.toml b/providers/cheaperinference/models/gpt-5.2-codex.toml
new file mode 100644
index 00000000000..4755e4c25b5
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.2-codex.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.75/$14.0 per 1M).
+base_model = "openai/gpt-5.2-codex"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh"]
+
+[cost]
+input = 1.225
+output = 9.8
+cache_read = 0.1225
+cache_write = 1.225
+
+[limit]
+context = 256000
+output = 65536
diff --git a/providers/cheaperinference/models/gpt-5.4-mini.toml b/providers/cheaperinference/models/gpt-5.4-mini.toml
new file mode 100644
index 00000000000..e75899b3609
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.4-mini.toml
@@ -0,0 +1,15 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.75/$4.5 per 1M).
+base_model = "openai/gpt-5.4-mini"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 0.525
+output = 3.15
+cache_read = 0.0525
+cache_write = 0.525
diff --git a/providers/cheaperinference/models/gpt-5.4-nano.toml b/providers/cheaperinference/models/gpt-5.4-nano.toml
new file mode 100644
index 00000000000..35272eca19d
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.4-nano.toml
@@ -0,0 +1,19 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# "black" twice and NOIMAGE twice across 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 19.43% off the model maker's list price ($0.2/$1.25 per 1M).
+base_model = "openai/gpt-5.4-nano"
+attachment = false
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 0.161137
+output = 1.007109
+cache_read = 0.016114
+cache_write = 0.161137
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/gpt-5.4.toml b/providers/cheaperinference/models/gpt-5.4.toml
new file mode 100644
index 00000000000..c8045ce6fb1
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.4.toml
@@ -0,0 +1,19 @@
+# Image input verified on this host: 3 of 4 colour probes exact and the fourth called "orange" for
+# yellow (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$15.0 per 1M).
+base_model = "openai/gpt-5.4"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 1.75
+output = 10.5
+cache_read = 0.175
+cache_write = 1.75
+
+[limit]
+context = 1000000
+output = 131072
diff --git a/providers/cheaperinference/models/gpt-5.5-pro.toml b/providers/cheaperinference/models/gpt-5.5-pro.toml
new file mode 100644
index 00000000000..b26a4104979
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.5-pro.toml
@@ -0,0 +1,25 @@
+# Image input verified on this host through the /v1/responses layer this route requires: a 32x32
+# solid-colour PNG as a data: URI was named correctly on 4 of 5 attempts, the fifth a transient 502
+# (probed 2026-09-17).
+# /v1/chat/completions returns 400 "Pro reasoning mode requires the /v1/responses endpoint", and /v1/responses
+# answers, so this route is responses-shaped. There effort medium and high are accepted, low returns 400 and
+# xhigh fails upstream (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($30.0/$180.0 per 1M).
+base_model = "openai/gpt-5.5-pro"
+
+[[reasoning_options]]
+type = "effort"
+values = ["medium", "high"]
+
+[cost]
+input = 21.0
+output = 126.0
+cache_read = 2.55
+cache_write = 21.0
+
+[limit]
+context = 1000000
+
+[provider]
+shape = "responses"
diff --git a/providers/cheaperinference/models/gpt-5.5.toml b/providers/cheaperinference/models/gpt-5.5.toml
new file mode 100644
index 00000000000..4aa5dcf0a60
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.5.toml
@@ -0,0 +1,18 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($5.0/$30.0 per 1M).
+base_model = "openai/gpt-5.5"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 3.5
+output = 21.0
+cache_read = 0.35
+cache_write = 3.5
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/gpt-5.6-luna.toml b/providers/cheaperinference/models/gpt-5.6-luna.toml
new file mode 100644
index 00000000000..f97d8800b7d
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.6-luna.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 60.00% off the model maker's list price ($0.2/$1.2 per 1M).
+base_model = "openai/gpt-5.6-luna"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 0.08
+output = 0.48
+cache_read = 0.008
+cache_write = 0.1
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 0.16
+output = 0.72
+cache_read = 0.016
+cache_write = 0.2
diff --git a/providers/cheaperinference/models/gpt-5.6-sol.toml b/providers/cheaperinference/models/gpt-5.6-sol.toml
new file mode 100644
index 00000000000..bf26e1ee568
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.6-sol.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 50.00% off the model maker's list price ($2.0/$10.0 per 1M).
+base_model = "openai/gpt-5.6-sol"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 1.0
+output = 5.0
+cache_read = 0.1
+cache_write = 1.25
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 2.0
+output = 7.5
+cache_read = 0.2
+cache_write = 2.5
diff --git a/providers/cheaperinference/models/gpt-5.6-terra.toml b/providers/cheaperinference/models/gpt-5.6-terra.toml
new file mode 100644
index 00000000000..fd948ab09fc
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-5.6-terra.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 60.00% off the model maker's list price ($2.0/$12.0 per 1M).
+base_model = "openai/gpt-5.6-terra"
+
+[[reasoning_options]]
+type = "effort"
+values = ["none", "low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 0.8
+output = 4.8
+cache_read = 0.08
+cache_write = 1.0
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 1.6
+output = 7.2
+cache_read = 0.16
+cache_write = 2.0
diff --git a/providers/cheaperinference/models/gpt-6-astra.toml b/providers/cheaperinference/models/gpt-6-astra.toml
new file mode 100644
index 00000000000..6f8f65529fe
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-6-astra.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($10.0/$50.0 per 1M).
+base_model = "openai/gpt-6-astra"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high", "xhigh", "max"]
+
+[cost]
+input = 7.0
+output = 35.0
+cache_read = 0.7
+cache_write = 8.5
+
+[[cost.tiers]]
+tier = { type = "context", size = 272_000 }
+input = 14.0
+output = 52.5
+cache_read = 1.4
+cache_write = 17.0
diff --git a/providers/cheaperinference/models/gpt-oss-120b.toml b/providers/cheaperinference/models/gpt-oss-120b.toml
new file mode 100644
index 00000000000..1ee0cbfe40c
--- /dev/null
+++ b/providers/cheaperinference/models/gpt-oss-120b.toml
@@ -0,0 +1,18 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 60.00% off the model maker's list price ($0.1/$0.5 per 1M).
+base_model = "openai/gpt-oss-120b"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 0.04
+output = 0.2
+cache_read = 0.0085
+cache_write = 0.04
diff --git a/providers/cheaperinference/models/grok-4.5.toml b/providers/cheaperinference/models/grok-4.5.toml
new file mode 100644
index 00000000000..6ca9c1491a8
--- /dev/null
+++ b/providers/cheaperinference/models/grok-4.5.toml
@@ -0,0 +1,21 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.0/$6.0 per 1M).
+base_model = "xai/grok-4.5"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
+[cost]
+input = 1.4
+output = 4.2
+cache_read = 0.17
+cache_write = 1.4
+
+[limit]
+output = 128000
diff --git a/providers/cheaperinference/models/kimi-k3.toml b/providers/cheaperinference/models/kimi-k3.toml
new file mode 100644
index 00000000000..9d11d32f6ca
--- /dev/null
+++ b/providers/cheaperinference/models/kimi-k3.toml
@@ -0,0 +1,25 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($3.0/$15.0 per 1M).
+base_model = "moonshotai/kimi-k3"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "high", "max"]
+
+[cost]
+input = 2.1
+output = 10.5
+cache_read = 0.239969
+cache_write = 2.1
+
+[limit]
+context = 1000000
diff --git a/providers/cheaperinference/models/minimax-m2.7.toml b/providers/cheaperinference/models/minimax-m2.7.toml
new file mode 100644
index 00000000000..0654c123111
--- /dev/null
+++ b/providers/cheaperinference/models/minimax-m2.7.toml
@@ -0,0 +1,22 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# it answered NOIMAGE on 4 of 4 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3/$1.2 per 1M).
+base_model = "minimax/MiniMax-M2.7"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.21
+output = 0.84
+cache_read = 0.0255
+cache_write = 0.255
+
+[limit]
+context = 198000
+output = 32768
diff --git a/providers/cheaperinference/models/muse-spark-1.2.toml b/providers/cheaperinference/models/muse-spark-1.2.toml
new file mode 100644
index 00000000000..e1d025ea6a4
--- /dev/null
+++ b/providers/cheaperinference/models/muse-spark-1.2.toml
@@ -0,0 +1,27 @@
+# No usable image input on this host: the request is accepted but the image never reaches the model -
+# NOIMAGE or an empty message on 6 of 6 colour probes (probed 2026-09-17).
+# Reasoning is returned in a reasoning_details field on the message: present on 3 of 3 untruncated probes
+# (runs cut short by max_tokens carry none), so the side channel is declared below (probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($1.25/$4.25 per 1M).
+base_model = "meta/muse-spark-1.2"
+attachment = false
+
+[interleaved]
+field = "reasoning_details"
+
+[[reasoning_options]]
+type = "effort"
+values = ["minimal", "low", "medium", "high", "xhigh"]
+
+[cost]
+input = 0.875
+output = 2.975
+cache_read = 0.105
+cache_write = 0.875
+
+[limit]
+output = 943718
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/qwen-3-8-27b.toml b/providers/cheaperinference/models/qwen-3-8-27b.toml
new file mode 100644
index 00000000000..c3879cd149a
--- /dev/null
+++ b/providers/cheaperinference/models/qwen-3-8-27b.toml
@@ -0,0 +1,25 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.214/$2.55 per 1M).
+base_model = "alibaba/qwen3.8-27b"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "xhigh"]
+
+[cost]
+input = 0.1498
+output = 1.785
+cache_read = 0.01819
+cache_write = 0.1498
+
+[limit]
+output = 65536
diff --git a/providers/cheaperinference/models/qwen-3-8-max.toml b/providers/cheaperinference/models/qwen-3-8-max.toml
new file mode 100644
index 00000000000..dc324f00d5f
--- /dev/null
+++ b/providers/cheaperinference/models/qwen-3-8-max.toml
@@ -0,0 +1,17 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Reasoning is always on and takes no caller control: reasoning_content comes back on every request, including
+# with reasoning.enabled = false and with reasoning.effort = "none" (6 of 6 probed 2026-09-16).
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($2.5/$7.5 per 1M).
+base_model = "alibaba/qwen3.8-max"
+reasoning_options = []
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.75
+output = 5.25
+cache_read = 0.2125
+cache_write = 2.125
diff --git a/providers/cheaperinference/models/qwen3-5-35b-a3b.toml b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
new file mode 100644
index 00000000000..2e609965597
--- /dev/null
+++ b/providers/cheaperinference/models/qwen3-5-35b-a3b.toml
@@ -0,0 +1,21 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.3125/$1.25 per 1M).
+base_model = "alibaba/qwen3.5-35b-a3b"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.21875
+output = 0.875
+cache_read = 0.026562
+cache_write = 0.21875
+
+[limit]
+context = 256000
diff --git a/providers/cheaperinference/models/qwen3-6-35b-a3b.toml b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
new file mode 100644
index 00000000000..7a047967e25
--- /dev/null
+++ b/providers/cheaperinference/models/qwen3-6-35b-a3b.toml
@@ -0,0 +1,25 @@
+# No image input on this host: an image_url part carrying a 32x32 PNG as a data: URI is rejected with
+# 400 on 6 of 6 requests while a text-only control answers 200 (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 30.00% off the model maker's list price ($0.1/$1.0 per 1M).
+base_model = "alibaba/qwen3.6-35b-a3b"
+attachment = false
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.07
+output = 0.7
+cache_read = 0.0085
+cache_write = 0.07
+
+[limit]
+output = 262144
+
+[modalities]
+input = ["text"]
diff --git a/providers/cheaperinference/models/qwen3.6-27b.toml b/providers/cheaperinference/models/qwen3.6-27b.toml
new file mode 100644
index 00000000000..89568257521
--- /dev/null
+++ b/providers/cheaperinference/models/qwen3.6-27b.toml
@@ -0,0 +1,22 @@
+# Image input verified on this host: a 32x32 solid-colour PNG as a data: URI was named correctly on
+# 4 of 4 colour probes (probed 2026-09-17).
+# Toggle: reasoning.enabled = true|false
+# Rates: GET https://api.cheaperinference.com/v1/models (pricing_version sha256:6d2520b9…, checked 2026-09-17T07:00:28Z)
+# Discounted marketplace rate, 55.24% off the model maker's list price ($0.6/$3.6 per 1M).
+base_model = "alibaba/qwen3.6-27b"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "toggle"
+
+[cost]
+input = 0.241706
+output = 1.611374
+cache_read = 0.024171
+cache_write = 0.241706
+
+[limit]
+context = 256000
+output = 32768
diff --git a/providers/cheaperinference/provider.toml b/providers/cheaperinference/provider.toml
new file mode 100644
index 00000000000..b3bc7116068
--- /dev/null
+++ b/providers/cheaperinference/provider.toml
@@ -0,0 +1,5 @@
+name = "CheaperInference"
+env = ["CHEAPERINFERENCE_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+api = "https://api.cheaperinference.com/v1"
+doc = "https://cheaperinference.com/docs"