From de8951b45e16f554ec1fb95673998b0820b83be0 Mon Sep 17 00:00:00 2001 From: Dawncoral <123850472@qq.com> Date: Sun, 13 Sep 2026 18:39:09 +0800 Subject: [PATCH 1/3] Add Wallaby provider (wallaby/kimi-k3) --- providers/wallaby/logo.svg | 3 +++ providers/wallaby/models/moonshotai/kimi-k3.toml | 6 ++++++ providers/wallaby/provider.toml | 5 +++++ 3 files changed, 14 insertions(+) create mode 100644 providers/wallaby/logo.svg create mode 100644 providers/wallaby/models/moonshotai/kimi-k3.toml create mode 100644 providers/wallaby/provider.toml diff --git a/providers/wallaby/logo.svg b/providers/wallaby/logo.svg new file mode 100644 index 00000000000..42950154749 --- /dev/null +++ b/providers/wallaby/logo.svg @@ -0,0 +1,3 @@ + + + diff --git a/providers/wallaby/models/moonshotai/kimi-k3.toml b/providers/wallaby/models/moonshotai/kimi-k3.toml new file mode 100644 index 00000000000..4fbe6eb5661 --- /dev/null +++ b/providers/wallaby/models/moonshotai/kimi-k3.toml @@ -0,0 +1,6 @@ +base_model = "moonshotai/kimi-k3" + +[cost] +input = 2.70 +output = 13.50 +cache_read = 0.27 diff --git a/providers/wallaby/provider.toml b/providers/wallaby/provider.toml new file mode 100644 index 00000000000..a73881791ac --- /dev/null +++ b/providers/wallaby/provider.toml @@ -0,0 +1,5 @@ +name = "Wallaby" +npm = "@ai-sdk/openai-compatible" +env = ["WALLABY_API_KEY"] +api = "https://api.wallabytoken.com/v1" +doc = "https://www.wallabytoken.com/docs" From eb74ab8174bff7e2cad58844015cb17cf16ad77c Mon Sep 17 00:00:00 2001 From: Dawncoral <123850472@qq.com> Date: Mon, 14 Sep 2026 17:07:24 +0800 Subject: [PATCH 2/3] Declare host-verified reasoning effort for wallaby/kimi-k3 reasoning_effort low/medium/high verified effective on this host (2026-09-14): graded reduction in reasoning tokens. none/minimal are accepted but do not disable reasoning, and omitting the parameter runs full effort, so declare effort values without none and no toggle. Also restore interleaved reasoning_content and align doc URL. --- providers/wallaby/models/moonshotai/kimi-k3.toml | 9 +++++++++ providers/wallaby/provider.toml | 6 +++++- 2 files changed, 14 insertions(+), 1 deletion(-) diff --git a/providers/wallaby/models/moonshotai/kimi-k3.toml b/providers/wallaby/models/moonshotai/kimi-k3.toml index 4fbe6eb5661..0ed9df7c9c8 100644 --- a/providers/wallaby/models/moonshotai/kimi-k3.toml +++ b/providers/wallaby/models/moonshotai/kimi-k3.toml @@ -1,4 +1,13 @@ +# Effort: reasoning_effort = low|medium|high (verified on this host 2026-09-14: +# graded reduction in reasoning tokens; "none"/"minimal" are accepted but do not +# disable reasoning; omitting the parameter runs full effort — hence no "none" +# in values and no toggle). +# Pricing: https://wallabytoken.com/pricing.json (accessed 2026-09-14), USD per 1M tokens. base_model = "moonshotai/kimi-k3" +reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }] + +[interleaved] +field = "reasoning_content" [cost] input = 2.70 diff --git a/providers/wallaby/provider.toml b/providers/wallaby/provider.toml index a73881791ac..ffa9498fe24 100644 --- a/providers/wallaby/provider.toml +++ b/providers/wallaby/provider.toml @@ -1,5 +1,9 @@ +# OpenAI-compatible API focused on Kimi K3. Reasoning effort is adjustable via +# `reasoning_effort` (low/medium/high) and reasoning streams in the +# `reasoning_content` field. `temperature` is accepted but normalized to 1. +# https://wallabytoken.com/docs (accessed 2026-09-14) name = "Wallaby" npm = "@ai-sdk/openai-compatible" env = ["WALLABY_API_KEY"] api = "https://api.wallabytoken.com/v1" -doc = "https://www.wallabytoken.com/docs" +doc = "https://wallabytoken.com/docs" From 1136765868c4ea609c2136747d95cf3a395664ed Mon Sep 17 00:00:00 2001 From: Dawncoral Date: Mon, 14 Sep 2026 19:07:17 +0800 Subject: [PATCH 3/3] Address review: unfixed logo size, add host-verified max effort tier - logo.svg: drop fixed width/height, keep square viewBox + currentColor (merge blocker for new providers) - reasoning_options: re-probed host 2026-09-14, same-prompt median reasoning tokens over 6 runs per level: low 61 < medium 85 < high 105 < max 123. max is accepted and highest; medium is a distinct graded intermediate tier on this wire. No wire path disables reasoning (none/minimal still reason; reasoning.enabled=false and thinking.type=disabled ignored), so no toggle and no none value. --- providers/wallaby/logo.svg | 2 +- providers/wallaby/models/moonshotai/kimi-k3.toml | 13 ++++++++----- 2 files changed, 9 insertions(+), 6 deletions(-) diff --git a/providers/wallaby/logo.svg b/providers/wallaby/logo.svg index 42950154749..2eddcadc840 100644 --- a/providers/wallaby/logo.svg +++ b/providers/wallaby/logo.svg @@ -1,3 +1,3 @@ - + diff --git a/providers/wallaby/models/moonshotai/kimi-k3.toml b/providers/wallaby/models/moonshotai/kimi-k3.toml index 0ed9df7c9c8..ed5e6e18fe3 100644 --- a/providers/wallaby/models/moonshotai/kimi-k3.toml +++ b/providers/wallaby/models/moonshotai/kimi-k3.toml @@ -1,10 +1,13 @@ -# Effort: reasoning_effort = low|medium|high (verified on this host 2026-09-14: -# graded reduction in reasoning tokens; "none"/"minimal" are accepted but do not -# disable reasoning; omitting the parameter runs full effort — hence no "none" -# in values and no toggle). +# Effort: reasoning_effort = low|medium|high|max (verified on this host +# 2026-09-14; same-prompt median reasoning tokens over 6 runs per level: +# low 61 < medium 85 < high 105 < max 123 — four graded tiers, "max" accepted +# and highest, "medium" a distinct intermediate tier). No wire path disables +# reasoning: "none"/"minimal" are accepted but still reason, and +# reasoning.enabled=false / thinking.type="disabled" are ignored — hence no +# "none" in values and no toggle. # Pricing: https://wallabytoken.com/pricing.json (accessed 2026-09-14), USD per 1M tokens. base_model = "moonshotai/kimi-k3" -reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }] +reasoning_options = [{ type = "effort", values = ["low", "medium", "high", "max"] }] [interleaved] field = "reasoning_content"