diff --git a/providers/wallaby/logo.svg b/providers/wallaby/logo.svg new file mode 100644 index 00000000000..2eddcadc840 --- /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..ed5e6e18fe3 --- /dev/null +++ b/providers/wallaby/models/moonshotai/kimi-k3.toml @@ -0,0 +1,18 @@ +# 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", "max"] }] + +[interleaved] +field = "reasoning_content" + +[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..ffa9498fe24 --- /dev/null +++ b/providers/wallaby/provider.toml @@ -0,0 +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://wallabytoken.com/docs"