feat: add kimi-k2.6-code-preview model support

[bzqzheng]

Issue for this PR

Relates to #22408

Type of change

• New feature
• Bug fix
• Refactor / code improvement
• Documentation

What does this PR do?

Adds client-side support for the kimi-k2.6-code-preview model so users who configure it manually (e.g. via opencode.json) receive the correct provider transforms.

Specific changes:

• packages/opencode/src/provider/transform.ts:
  • Added kimi-k2.6-code-preview to topP() so the model receives topP: 0.95.
  • Added kimi-k2.6-code-preview to the Anthropic SDK thinking logic so thinking is enabled by default.
  • temperature() already handles kimi-k2.6-code-preview correctly via the existing k2. pattern match.
• packages/opencode/test/tool/fixtures/models-api.json:
  • Added kimi-k2.6-code-preview fixture entries for the kimi-for-coding and opencode providers.
• packages/console/app/src/routes/workspace/[id]/usage/graph-section.tsx:
  • Added a color mapping for kimi-k2.6-code-preview in usage graphs.

The base kimi-k2.6 model is intentionally left for a follow-up PR since its provider transforms need separate validation.

How did you verify your code works?

• Ran bun test test/provider/ and bun test test/tool/ from packages/opencode — all tests pass.
• The pre-push typecheck hook also passes across the workspace.

Screenshots / recordings

N/A — no UI behavior changes beyond the usage graph color mapping.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[Acters]

kimi's k2.6 and k2.6-code-preview are completely different models. It seems that you are mixing them up as the same thing.
Also, why are there so many changes made?
I hope this is not vibe coded slop.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍

[Acters]

These new changes are great improvement.
Can you change it to add the normal k2.6 model alongside the k2.6-code-preview?
Also, the maintainers might not merge this soon because it is a preview model. I wonder if people are hesitant to add the model until there is a full release.
BTW, I just tested this a minute ago as I was writing this and there is an id: k2.6-code model available.
Its output seems to be better than the preview model.
Want to weigh in on this?

[bzqzheng]

kimi's k2.6 and k2.6-code-preview are completely different models. It seems that you are mixing them up as the same thing. Also, why are there so many changes made? I hope this is not vibe coded slop.

re: k2.6 vs. k2.6-code-preview
updated packages/opencode/src/provider/transform.ts to use the explicit model ID kimi-k2.6-code-preview.

re: unrelated changes
upstream commit history messed up and got accidentally bundled into the branch; already rebased. also meant to open the PR as draft before the gh checks pass.

[bzqzheng]

These new changes are great improvement. Can you change it to add the normal k2.6 model alongside the k2.6-code-preview? Also, the maintainers might not merge this soon because it is a preview model. I wonder if people are hesitant to add the model until there is a full release. BTW, I just tested this a minute ago as I was writing this and there is an id: k2.6-code model available. Its output seems to be better than the preview model. Want to weigh in on this?

how did you discover id k2.6-code?

[Acters]

These new changes are great improvement. Can you change it to add the normal k2.6 model alongside the k2.6-code-preview? Also, the maintainers might not merge this soon because it is a preview model. I wonder if people are hesitant to add the model until there is a full release. BTW, I just tested this a minute ago as I was writing this and there is an id: k2.6-code model available. Its output seems to be better than the preview model. Want to weigh in on this?

how did you discover id k2.6-code?

As I was writing it, a thought popped in my head "wait its a preview, that would mean there might be a non-preview model"

I edited my opencode.json with this:

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "kimi-for-coding": {
      "models": {
        "Kimi K2.6": {
          "id": "kimi-k2.6",
          "reasoning": true,
          "attachment": false,
          "limit": {
            "context": 262144,
            "output": 32768
          },
          "modalities": {
            "input": ["text", "image", "video"],
            "output": ["text"]
          },
          "options": {
            "interleaved": {
              "field": "reasoning_content"
            }
          }
        },
        "Kimi K2.6 code preview": {
          "id": "kimi-k2.6-code-preview",
          "reasoning": true,
          "attachment": false,
          "limit": {
            "context": 262144,
            "output": 32768
          },
          "modalities": {
            "input": ["text", "image", "video"],
            "output": ["text"]
          },
          "options": {
            "interleaved": {
              "field": "reasoning_content"
            }
          }
        },
        "Kimi K2.6 code": {
          "id": "kimi-k2.6-code",
          "reasoning": true,
          "attachment": false,
          "limit": {
            "context": 262144,
            "output": 32768
          },
          "modalities": {
            "input": ["text", "image", "video"],
            "output": ["text"]
          },
          "options": {
            "interleaved": {
              "field": "reasoning_content"
            }
          }
        }
      }
    }
  }
}

It worked, so I gather there is some stuff happening in the backend for Kimi code that is not public yet, haha.
On the other hand, the backend might just be capturing this ID kimi-k2.6-code and directing it to the preview model.

[Acters]

btw, I noticed that we have been doing non-thinking so far. Is it possible that you can do a thinking enabled variant?