From 917b558663a6f17e93abb7d4f3d5fb36414ed901 Mon Sep 17 00:00:00 2001 From: Jonathan Campbell Evans Date: Fri, 3 Apr 2026 12:48:07 -0700 Subject: [PATCH 1/2] feat: add FirePass kimi-k2p5-turbo router model --- packages/types/src/providers/fireworks.ts | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/packages/types/src/providers/fireworks.ts b/packages/types/src/providers/fireworks.ts index c9017c54cde..87a20c4342d 100644 --- a/packages/types/src/providers/fireworks.ts +++ b/packages/types/src/providers/fireworks.ts @@ -22,6 +22,7 @@ export type FireworksModelId = | "accounts/fireworks/models/llama-v3p3-70b-instruct" | "accounts/fireworks/models/llama4-maverick-instruct-basic" | "accounts/fireworks/models/llama4-scout-instruct-basic" + | "accounts/fireworks/routers/kimi-k2p5-turbo" export const fireworksDefaultModelId: FireworksModelId = "accounts/fireworks/models/kimi-k2-instruct-0905" @@ -72,6 +73,17 @@ export const fireworksModels = { description: "Kimi K2.5 is Moonshot AI's flagship agentic model and a new SOTA open model. It unifies vision and text, thinking and non-thinking modes, and single-agent and multi-agent execution into one model. Fireworks enables users to control the reasoning behavior and inspect its reasoning history for greater transparency.", }, + "accounts/fireworks/routers/kimi-k2p5-turbo": { + maxTokens: 16384, + contextWindow: 262144, + supportsImages: true, + supportsPromptCache: true, + inputPrice: 0, + outputPrice: 0, + cacheReadsPrice: 0, + description: + "Kimi K2.5 Turbo router - same configuration as kimi-k2p5 but with improved performance through Fireworks routing (FirePass - https://docs.fireworks.ai/firepass).", + }, "accounts/fireworks/models/minimax-m2": { maxTokens: 4096, contextWindow: 204800, From eb51e1117cfdc56c02ffb78dbb30c823ffd193e0 Mon Sep 17 00:00:00 2001 From: Jonathan Evans Date: Mon, 6 Apr 2026 14:16:57 -0700 Subject: [PATCH 2/2] fix: disable prompt caching for kimi-k2p5-turbo Roo Code passes Anthropic-style cache_control blocks in OpenAI-compatible API calls, which causes validation errors when using Fireworks as an OpenAI-compatible provider. Disable prompt caching until this is fixed. --- packages/types/src/providers/fireworks.ts | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/packages/types/src/providers/fireworks.ts b/packages/types/src/providers/fireworks.ts index 87a20c4342d..2489aeff6bb 100644 --- a/packages/types/src/providers/fireworks.ts +++ b/packages/types/src/providers/fireworks.ts @@ -77,10 +77,9 @@ export const fireworksModels = { maxTokens: 16384, contextWindow: 262144, supportsImages: true, - supportsPromptCache: true, + supportsPromptCache: false, inputPrice: 0, outputPrice: 0, - cacheReadsPrice: 0, description: "Kimi K2.5 Turbo router - same configuration as kimi-k2p5 but with improved performance through Fireworks routing (FirePass - https://docs.fireworks.ai/firepass).", },