fix(reasoning): use adaptive thinking for Claude 4.7+ models

[Raymond8196]

What this PR does

Before this PR:

Claude Opus 4.7 (claude-opus-4-7) fails with HTTP 400 when reasoning effort is set to any non-default value (low, medium, high). The error message: "thinking.type.enabled" is not supported for this model. Use "thinking.type.adaptive" and "output_config.effort" to control thinking behavior.

After this PR:

Claude Opus 4.7 correctly uses thinking: { type: "adaptive" } with the effort parameter, matching the Anthropic API requirements.

Fixes #14805
Fixes #14821

Why we need it and why it was done in this way

The root cause is that isClaude46SeriesModel() in utils.ts used a regex that only matched 4-6/4.6. Claude Opus 4.7 failed this check and fell through to the old enabled + budgetTokens code path, which Opus 4.7 does not support (returns HTTP 400).

The fix expands the regex from 4[.-]6 to 4[.-](?:[6-9]|\d{2,}) to match Claude 4.6+ models. This is the minimal change that fixes the issue and is forward-compatible with future 4.8, 4.9 releases, which are expected to continue using adaptive thinking (Anthropic is deprecating the enabled + budgetTokens approach).

The following tradeoffs were made:

• Chose to match all 4.6+ versions rather than adding a separate isClaude47SeriesModel() function, since the adaptive thinking API is the same for both 4.6 and 4.7+.

The following alternatives were considered:

• Adding a separate isClaude47SeriesModel() check — rejected as unnecessary duplication since both 4.6 and 4.7 use the same adaptive thinking API.

Note: Some proxy APIs may automatically convert enabled + budget_tokens to adaptive parameters, masking this issue. Users connecting directly to the Anthropic API or AWS Bedrock are affected.

Breaking changes

None.

Special notes for your reviewer

All 4 call sites of isClaude46SeriesModel() benefit from this change:

1. reasoning.ts:732 — Uses adaptive + effort instead of enabled + budgetTokens
2. reasoning.ts:956 — Bedrock adaptive thinking (same logic)
3. modelParameters.ts:146 — Skips budget subtraction from maxTokens
4. reasoning.ts:136 — UI displays claude46 effort options (includes xhigh)

Checklist

• PR: The PR description is expressive enough and will help future contributors
• Code: Write code that humans can understand and Keep it simple
• Refactor: You have left the code cleaner than you found it (Boy Scout Rule)
• Upgrade: Impact of this change on upgrade flows was considered and addressed if required
• Documentation: A user-guide update was considered and is present (link) or not required. Check this only when the PR introduces or changes a user-facing feature or behavior.
• Self-review: I have reviewed my own code (e.g., via /gh-pr-review, gh pr diff, or GitHub UI) before requesting review from others

Release note

Fix Claude Opus 4.7 returning HTTP 400 when reasoning effort is set to non-default values. The model now correctly uses adaptive thinking parameters instead of the unsupported extended thinking format.

[ousugo]

发现 1 个需要补强的点：这次 4.6+ 识别修复缺少针对 Claude 4.7 / Bedrock 4.7 的回归测试，建议补上真实 model id 的断言以锁住修复路径。

[DeJeune]

may be fixed: #14349

[cloudy-sfu]

Confirmed the fix #14349 works.