feat(zenifra): add hosted model catalog

packages/core/test/generate.test.ts

@@ -45,6 +45,160 @@ describe("catalog generation", () => {
     });
   });
 
+  test("Zenifra model IDs preserve the upstream namespace", async () => {
+    const root = path.join(import.meta.dirname, "..", "..", "..");
+    const providers = await generate(path.join(root, "providers"));
+    const modelIDs = Object.keys(providers.zenifra?.models ?? {});
+
+    expect(modelIDs).not.toEqual([]);
+    expect(modelIDs.every((modelID) => modelID.startsWith("zenifra/"))).toBe(true);
+  });
+
+  test("Zenifra exposes only native reasoning controls", async () => {
+    const root = path.join(import.meta.dirname, "..", "..", "..");
+    const providers = await generate(path.join(root, "providers"));
+    const models = providers.zenifra?.models ?? {};
+
+    const actual = Object.fromEntries(
+      Object.entries(models).map(([modelID, model]) => [
+        modelID,
+        model.reasoning_options,
+      ]),
+    );
+
+    expect(actual).toEqual({
+      "zenifra/deepseek-v4-flash-0731": [
+        { type: "toggle" },
+        { type: "effort", values: ["low", "high", "max"] },
+      ],
+      "zenifra/deepseek-v4-pro": [
+        { type: "toggle" },
+        { type: "effort", values: ["high", "max"] },
+      ],
+      "zenifra/glm-5.1": [{ type: "toggle" }],
+      "zenifra/glm-5.2": [
+        { type: "toggle" },
+        { type: "effort", values: ["high", "max"] },
+      ],
+      "zenifra/kimi-k2.5": [{ type: "toggle" }],
+      "zenifra/kimi-k2.7-code": [],
+      "zenifra/kimi-k3": [
+        { type: "effort", values: ["low", "high", "max"] },
+      ],
+      "zenifra/qwen3.6-flash": [{ type: "toggle" }],
+      "zenifra/qwen3.6-plus": [{ type: "toggle" }],
+      "zenifra/qwen3.7-max": [{ type: "toggle" }],
+      "zenifra/qwen3.7-plus": [{ type: "toggle" }],
+      "zenifra/qwen3.8-27b": [
+        { type: "toggle" },
+        { type: "effort", values: ["low", "medium", "xhigh"] },
+      ],
+      "zenifra/qwen3.8-flash": [
+        { type: "toggle" },
+        { type: "effort", values: ["low", "medium", "xhigh"] },
+      ],
+      "zenifra/qwen3.8-max": [
+        { type: "toggle" },
+        { type: "effort", values: ["low", "medium", "xhigh"] },
+      ],
+    });
+  });
+
+  test("Zenifra attachment support matches its served input modalities", async () => {
+    const root = path.join(import.meta.dirname, "..", "..", "..");
+    const providers = await generate(path.join(root, "providers"));
+    const models = providers.zenifra?.models ?? {};
+
+    const actual = Object.fromEntries(
+      Object.entries(models).map(([modelID, model]) => [
+        modelID,
+        model.attachment,
+      ]),
+    );
+
+    expect(actual).toEqual({
+      "zenifra/deepseek-v4-flash-0731": false,
+      "zenifra/deepseek-v4-pro": false,
+      "zenifra/glm-5.1": false,
+      "zenifra/glm-5.2": false,
+      "zenifra/kimi-k2.5": true,
+      "zenifra/kimi-k2.7-code": true,
+      "zenifra/kimi-k3": true,
+      "zenifra/qwen3.6-flash": true,
+      "zenifra/qwen3.6-plus": true,
+      "zenifra/qwen3.7-max": false,
+      "zenifra/qwen3.7-plus": true,
+      "zenifra/qwen3.8-27b": false,
+      "zenifra/qwen3.8-flash": true,
+      "zenifra/qwen3.8-max": true,
+    });
+  });
+
+  test("Zenifra serves current limits, modalities, and converted tier prices", async () => {
+    const root = path.join(import.meta.dirname, "..", "..", "..");
+    const providers = await generate(path.join(root, "providers"));
+    const models = providers.zenifra?.models ?? {};
+
+    expect(models["zenifra/glm-5.1"]?.limit).toEqual({
+      context: 202_745,
+      output: 131_072,
+    });
+    expect(models["zenifra/qwen3.6-flash"]?.limit).toEqual({
+      context: 991_808,
+      output: 65_536,
+    });
+    expect(models["zenifra/qwen3.6-plus"]?.limit).toEqual({
+      context: 991_808,
+      output: 65_536,
+    });
+    expect(models["zenifra/qwen3.8-27b"]?.modalities?.input).toEqual(["text"]);
+    expect(models["zenifra/qwen3.8-max"]?.modalities?.input).toEqual(["text", "image", "video"]);
+
+    expect(models["zenifra/glm-5.1"]?.cost).toEqual({
+      input: 1.144231,
+      output: 4.578846,
+      cache_read: 0.228846,
+      tiers: [{
+        tier: { type: "context", size: 32_001 },
+        input: 1.526923,
+        output: 5.340385,
+        cache_read: 0.305769,
+      }],
+    });
+    expect(models["zenifra/qwen3.6-flash"]?.cost).toEqual({
+      input: 0.163462,
+      output: 0.980769,
+      cache_read: 0.017308,
+      context_over_200k: {
+        input: 0.653846,
+        output: 3.923077,
+        cache_read: 0.065385,
+      },
+      tiers: [{
+        tier: { type: "context", size: 256_001 },
+        input: 0.653846,
+        output: 3.923077,
+        cache_read: 0.065385,
+      }],
+    });
+    expect(models["zenifra/qwen3.6-plus"]?.cost).toEqual({
+      input: 0.275,
+      output: 1.636538,
+      cache_read: 0.028846,
+      context_over_200k: {
+        input: 1.092308,
+        output: 6.540385,
+        cache_read: 0.111538,
+      },
+      tiers: [{
+        tier: { type: "context", size: 256_001 },
+        input: 1.092308,
+        output: 6.540385,
+        cache_read: 0.111538,
+      }],
+    });
+  });
+
   test("base_model can factor metadata without changing provider JSON", async () => {
     await withFixture(async (root) => {
       await write(root, "providers/direct/provider.toml", providerToml("Direct"));

providers/zenifra/models/zenifra/deepseek-v4-flash-0731.toml

@@ -0,0 +1,23 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Off-peak 14:00-00:00 UTC: BRL/1M input 1.10, output 3.30, cache read 0.14;
+# the models.dev cost schema does not represent time-of-day pricing.
+# Toggle: enable_thinking = true|false.
+# Effort: reasoning_effort = low|high|max.
+base_model = "deepseek/deepseek-v4-flash-0731"
+structured_output = false
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["low", "high", "max"] },
+]
+
+[limit]
+output = 393_216
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 0.423077
+output = 1.269231
+cache_read = 0.055769

providers/zenifra/models/zenifra/deepseek-v4-pro.toml

@@ -0,0 +1,23 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Toggle: enable_thinking = true|false.
+# Effort: reasoning_effort = high|max (GET /v1/models, 2026-09-22).
+# Off-peak 14:00-00:00 UTC: BRL/1M input 4.30, output 8.50, cache read 0.35;
+# the models.dev cost schema does not represent time-of-day pricing.
+base_model = "deepseek/deepseek-v4-pro-0813"
+structured_output = false
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["high", "max"] },
+]
+
+[limit]
+output = 393_216
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 1.634615
+output = 3.269231
+cache_read = 0.134615

providers/zenifra/models/zenifra/glm-5.1.toml

@@ -0,0 +1,23 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-22).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate).
+# No native reasoning_effort levels advertised; expose the model's thinking toggle only.
+# Toggle: enable_thinking = true|false (top-level chat/completions request).
+base_model = "zhipuai/glm-5.1"
+reasoning_options = [{ type = "toggle" }]
+
+[limit]
+context = 202_745
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 1.144231
+output = 4.578846
+cache_read = 0.228846
+
+[[cost.tiers]]
+tier = { type = "context", size = 32_001 }
+input = 1.526923
+output = 5.340385
+cache_read = 0.305769

providers/zenifra/models/zenifra/glm-5.2.toml

@@ -0,0 +1,21 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Toggle: enable_thinking = true|false.
+# Effort: reasoning_effort = high|max.
+base_model = "zhipuai/glm-5.2"
+structured_output = false
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["high", "max"] },
+]
+
+[limit]
+context = 1_048_576
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 1.096154
+output = 3.846154
+cache_read = 0.269231

providers/zenifra/models/zenifra/kimi-k2.5.toml

@@ -0,0 +1,20 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Capabilities: structured_outputs = false.
+# Toggle: enable_thinking = true|false; no native reasoning_effort levels.
+base_model = "moonshotai/kimi-k2.5"
+temperature = true
+structured_output = false
+reasoning_options = [{ type = "toggle" }]
+
+[limit]
+context = 229_376
+output = 16_384
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 0.576923
+output = 3.019231
+cache_read = 0.115385

providers/zenifra/models/zenifra/kimi-k2.7-code.toml

@@ -0,0 +1,18 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Reasoning is always on and no caller control is listed in supported_parameters.
+base_model = "moonshotai/kimi-k2.7-code"
+temperature = true
+reasoning_options = []
+
+[limit]
+context = 229_376
+output = 16_384
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 0.903846
+output = 3.730769
+cache_read = 0.067308

providers/zenifra/models/zenifra/kimi-k3.toml

@@ -0,0 +1,17 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Reasoning is always on; Zenifra advertises effort levels low|high|max.
+base_model = "moonshotai/kimi-k3"
+temperature = true
+reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }]
+
+[limit]
+output = 1_048_576
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 2.826923
+output = 14.132692
+cache_read = 0.288462

providers/zenifra/models/zenifra/qwen3.6-flash.toml

@@ -0,0 +1,23 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-22).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate).
+# No native reasoning_effort levels advertised; expose the model's thinking toggle only.
+# Toggle: enable_thinking = true|false (top-level chat/completions request).
+base_model = "alibaba/qwen3.6-flash"
+reasoning_options = [{ type = "toggle" }]
+
+[limit]
+context = 991_808
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 0.163462
+output = 0.980769
+cache_read = 0.017308
+
+[[cost.tiers]]
+tier = { type = "context", size = 256_001 }
+input = 0.653846
+output = 3.923077
+cache_read = 0.065385

providers/zenifra/models/zenifra/qwen3.6-plus.toml

@@ -0,0 +1,23 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-22).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate).
+# No native reasoning_effort levels advertised; expose the model's thinking toggle only.
+# Toggle: enable_thinking = true|false (top-level chat/completions request).
+base_model = "alibaba/qwen3.6-plus"
+reasoning_options = [{ type = "toggle" }]
+
+[limit]
+context = 991_808
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 0.275000
+output = 1.636538
+cache_read = 0.028846
+
+[[cost.tiers]]
+tier = { type = "context", size = 256_001 }
+input = 1.092308
+output = 6.540385
+cache_read = 0.111538

providers/zenifra/models/zenifra/qwen3.7-max.toml

@@ -0,0 +1,18 @@
+# Source: GET https://ai.zenifra.com/v1/models (accessed 2026-09-12).
+# Conversion: Zenifra prices are BRL; models.dev costs use BRL / 5.20 = USD (project rate, applied 2026-09-12).
+# Capabilities: structured_outputs = true; input_modalities = text.
+# Toggle: enable_thinking = true|false; no native reasoning_effort levels.
+base_model = "alibaba/qwen3.7-max"
+structured_output = true
+reasoning_options = [{ type = "toggle" }]
+
+[limit]
+output = 131_072
+
+[provider]
+shape = "completions"
+
+[cost]
+input = 1.634615
+output = 4.903846
+cache_read = 0.326923