fix(copilot): correct Claude thinking variants and context windows

[MoonFuji]

Summary

Two bugs fixed for Claude models served via GitHub Copilot:

1. Multi-level thinking budget variants (transform.ts)

Before: Only a single binary thinking variant with a hardcoded 4,000-token budget.

After: Four granular budget levels matching standard tiers:

Variant	thinking_budget
low	1,000 tokens
medium	4,000 tokens
high	10,000 tokens
max	16,000 tokens

Uses the Copilot SDK flat snake_case format (thinking_budget: number) as defined in openai-compatible-chat-options.ts — not the @ai-sdk/anthropic format.

2. Correct context windows (provider.ts)

Before: Context windows sourced from models.dev which contains stale/incorrect values (e.g. claude-opus-4.6 shown as ~128K).

After: Override table applied when m.api.npm === "@ai-sdk/github-copilot", based on official Anthropic documentation:

Model	Correct context
claude-opus-4.6, claude-sonnet-4.6	1,000,000 tokens
claude-opus-4.5, claude-sonnet-4.5, claude-haiku-4.5	200,000 tokens

Both dot-notation (4.6) and dash-notation (4-6) IDs are handled to cover API naming variations.

Files changed

• packages/opencode/src/provider/transform.ts
• packages/opencode/src/provider/provider.ts

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on the search results, I found two highly related PRs that may be duplicates or overlapping with PR #20319:

Related PRs:

1. PR fix(opencode): correct GitHub Copilot Claude context limits and thinking variants #20318 - "fix(opencode): correct GitHub Copilot Claude context limits and thinking variants"

   • fix(opencode): correct GitHub Copilot Claude context limits and thinking variants #20318
   • Why related: Appears to address the exact same issues - GitHub Copilot Claude context limits and thinking variants. This was likely opened just before or concurrently with PR fix(copilot): correct Claude thinking variants and context windows #20319.

2. PR fix(provider): add thinking level variants for GitHub Copilot Claude models #15167 - "fix(provider): add thinking level variants for GitHub Copilot Claude models"

   • fix(provider): add thinking level variants for GitHub Copilot Claude models #15167
   • Why related: Specifically addresses adding thinking level variants for GitHub Copilot Claude models, which is one of the key fixes in PR fix(copilot): correct Claude thinking variants and context windows #20319.

3. PR fix(provider): convert thinking.budgetTokens to snake_case for openai-compatible SDK #12544 - "fix(provider): convert thinking.budgetTokens to snake_case for openai-compatible SDK"

   • fix(provider): convert thinking.budgetTokens to snake_case for openai-compatible SDK #12544
   • Why related: Addresses the snake_case transformation for thinking_budget used in the Copilot SDK, relevant to the fix in transform.ts.

4. PR feat(provider): add adaptive thinking and 1M context support for Claude Opus 4.6 #12342 - "feat(provider): add adaptive thinking and 1M context support for Claude Opus 4.6"

   • feat(provider): add adaptive thinking and 1M context support for Claude Opus 4.6 #12342
   • Why related: Directly relevant to the context window fix for Claude Opus 4.6 (1M tokens).

Most critical: Check PR #20318 first - it appears to be the closest duplicate addressing the same fixes.

[rekram1-node]

U cant just override the context windows, these arent going to be accurate for most copilot users.

[github-actions]

This pull request has been automatically closed because it was not updated to meet our contributing guidelines within the 2-hour window.

Feel free to open a new pull request that follows our guidelines.