[ENHANCEMENT] Add Kimi K2.5 to Fireworks

[nakamorichi]

Problem (one or two sentences)

Add Kimi K2.5 support to Fireworks AI provider.

Context (who is affected and when)

.

Desired behavior (conceptual, not technical)

.

Constraints / preferences (optional)

.

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear context and impact

Roo Code Task Links (optional)

No response

Acceptance criteria (optional)

No response

Proposed approach (optional)

No response

Trade-offs / risks (optional)

No response

[roomote]

I would like to help with this issue, but I need some information about the Kimi K2.5 model to implement it correctly.

I found where Fireworks models are defined in the codebase (packages/types/src/providers/fireworks.ts). The existing Kimi K2 models (kimi-k2-instruct, kimi-k2-instruct-0905, kimi-k2-thinking) are already supported.

Could you please provide the following specifications for Kimi K2.5:

1. Fireworks Model ID - The exact model identifier (e.g., accounts/fireworks/models/kimi-k2p5-instruct)
2. Max output tokens - Maximum tokens the model can generate in a response
3. Context window size - Total context window in tokens
4. Image support - Does it support image inputs? (true/false)
5. Prompt cache support - Does it support prompt caching? (true/false)
6. Pricing - Input price and output price per million tokens (and cache reads price if applicable)
7. Special properties - Any special configurations like reasoning mode, default temperature, etc.
8. Description - A brief description of the model capabilities

Proposed implementation plan:

1. Add the new model ID to the FireworksModelId type union in packages/types/src/providers/fireworks.ts
2. Add the model configuration to the fireworksModels object with all the specifications above
3. Update the test file src/api/providers/__tests__/fireworks.spec.ts to include tests for the new model
4. Run tests to verify everything works correctly

Please mention @roomote in a comment with the model specifications, and I'll proceed with the implementation.

[nakamorichi]

@roomote Created PR: #11035

[roomote]

Thanks for creating the PR! I've reviewed PR #11035 and it looks like a solid implementation for adding Kimi K2.5 support. The model configuration includes all the key details (context window, pricing, image support, etc.).

One minor observation: the PR also changes the default Fireworks model from kimi-k2-instruct-0905 to kimi-k2p5. This might be intentional but could be a breaking change for users relying on the current default. Worth considering if this should be a separate discussion.

Thank you for the contribution!

[kavehsfv]

@roomote
Please add Kimi K2.5 to Moonshot Provider as it is the developer of this model.

[roomote]

I've seen your comment and will explore adding Kimi K2.5 to the Moonshot Provider. Since Moonshot AI is the developer of the Kimi models, this makes sense. I'll investigate the codebase structure and open a PR to share progress.

[roomote]

Opened a PR with my attempt at adding Kimi K2.5 to the Moonshot Provider: #11050

The model configuration includes vision support, prompt caching, and pricing details based on the Fireworks implementation. Feedback welcome!

[davidmoshal]

this is doable today, just:

a) use fireworks as the provider

b) enter this model in the model seach bar:
accounts/fireworks/models/kimi-k2p5

That will give you k2.5 at blazing speed, almost 200 tps:

• https://artificialanalysis.ai/models/kimi-k2-5/providers