fix(core): use decimal Claude output limits - #6735
Merged
Merged
Conversation
Collaborator
|
Thanks for the PR! Template looks good ✓ Problem: Observed bug with evidence. Linked issue #6734 provides a clear reproduction — Anthropic's API rejects Direction: Correctly aligned. The fix targets a real user-facing 400 error for anyone using Claude Opus 4.6-4.8 without an explicit Size: 1 production line + 7 test lines. Minimal and focused. Approach: Scope is exactly right. One line changed in production code, tests updated to match. No scope creep, no drive-by refactors. The inline Moving on to code review. 🔍 中文说明感谢贡献！ 模板完整 ✓ 问题：已观测到的 bug，有证据支持。 关联 issue #6734 提供了清晰的复现——Anthropic API 拒绝 方向：正确对齐。修复的是真实用户场景下的 400 错误——使用 Claude Opus 4.6-4.8 且不手动配置 规模：1 行生产代码 + 7 行测试。极简且聚焦。 方案：范围合理。生产代码只改了一行，测试同步更新。没有 scope creep，没有顺手重构。内联的 进入代码审查 🔍 — Qwen Code · qwen3.7-max |
Collaborator
Code ReviewThe diff is a single-line production change: the Claude Opus 4.6-4.8 output pattern switches from Independent proposal: I would have done exactly this — use an inline literal for the Claude-specific decimal limit rather than mutating the shared No issues found. The change is correct, minimal, and well-tested. The regex pattern Unit tests: 75/75 pass in Real-Scenario Testing (tmux)Verified the Before (installed build / main branch)After (this PR)All three Claude Opus models now correctly resolve to 128000. Sonnet 4.6 remains at 65536 (unchanged, correct). 中文说明代码审查diff 只改了一行生产代码：Claude Opus 4.6-4.8 的输出模式从 独立方案： 我会做完全一样的事情——用内联字面量表示 Claude 专属的十进制限制，而不是修改共享的 未发现问题。 改动正确、极简、测试覆盖充分。正则 单元测试： 真实场景测试（tmux）修复前后对比验证了 修复前（main 分支）：三个 Claude Opus 模型的 output 均为 131072（超出 Anthropic API 上限 128000）。 修复后（本 PR）：三个 Claude Opus 模型的 output 均为 128000（正确匹配 Anthropic API 上限）。Sonnet 4.6 保持 65536 不变。 — Qwen Code · qwen3.7-max |
Contributor
Code Coverage Summary
CLI Package - Full Text ReportCore Package - Full Text ReportFor detailed HTML reports, please see the 'coverage-reports-22.x-ubuntu-latest' artifact from the main CI run. |
Collaborator
|
This is a clean, well-evidenced bugfix. One line of production code changed, tests updated, before/after tmux output confirms the fix works exactly as described. The problem is real — Anthropic's API rejects I'd have written the same fix myself. Ships it. ✅ 中文说明这是一个干净、有充分证据的 bugfix。生产代码改了一行，测试同步更新，tmux 前后对比确认修复效果与描述一致。 问题是真实的——Anthropic API 拒绝 如果是我自己写，也会写出一模一样的修复。合并吧。✅ — Qwen Code · qwen3.7-max |
qwen-code-ci-bot
approved these changes
Jul 11, 2026
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
LGTM, looks ready to ship. ✅
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Reviewed — no blockers. Suggestions are inline.
Collaborator
✅ Local verification — merge referenceI built and ran the real test suite locally in an isolated git worktree checked out at the PR head ( TL;DR: the fix is correct, minimal and centralized. Claude Opus 4.6–4.8 now default to Anthropic's exact decimal ceiling Before / AfterRan the PR's updated specs against the pre-PR source (
Runtime evidencePrinted the values the CLI actually resolves at runtime. Opus 4.6–4.8 →
One policy point → propagates everywhere
Checks
Environment: macOS arm64 · isolated worktree at PR head Recommendation: LGTM — safe to merge. Single-line behavioral change ( 🇨🇳 中文版本（点击展开）✅ 本地验证 —— 合并参考我在一个隔离的 git worktree 中（检出到 PR head 结论： 该修复正确、最小化且集中。Claude Opus 4.6–4.8 现在默认使用 Anthropic 精确的十进制上限 前后对比将 PR 更新后的测试分别跑在合并前源码（ 运行时证据打印了 CLI 在运行时实际解析出的数值：Opus 4.6–4.8 → 单一策略点 → 全链路生效
检查项
环境：macOS arm64 · 检出到 PR head 建议：LGTM —— 可以安全合并。 单行行为变更（ |
wenshao
approved these changes
Jul 11, 2026
Collaborator
Author
|
Reviewed the remaining suggestion-only thread. I am not changing code for this PR because moving the inline |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this PR does
Updates the default
max_tokensfor Claude Opus 4.6, 4.7, and 4.8 from the binary131072value to Anthropic's exact API ceiling of128000. The sharedLIMITS['128k']value remains unchanged, so models such as GLM that accept131072are unaffected.Why it's needed
#6718 added the 1M context and 128K output rules for Opus 4.6-4.8, but reused the binary 128K constant. Anthropic-compatible endpoints reject
131072because their maximum is128000, causing a 400 for users who do not explicitly configuresamplingParams.max_tokens.Reviewer Test Plan
How to verify
Remove any explicit
samplingParams.max_tokensfrom a Claude Opus model configuration, then confirm thatclaude-opus-4-6,claude-opus-4-7, and provider-prefixed IDs such asvertex/claude-opus-4-8resolve to128000. Also confirm that an OpenAI-compatible request withoutmax_tokensinjects128000.Evidence (Before & After)
Tmux testing: before this change,
claude-opus-4-6was rejected withmax_tokens: 131072 > 128000. After this change, the same path with an isolatedQWEN_HOMEand no explicitmax_tokensreturnedTMUX_46_128000_OKwith exit code 0.Tested on
Environment (optional)
macOS arm64;
npm run dev; isolatedQWEN_HOME; qwen-code 0.19.9.Risk & Scope
claude-opus-4-8tmux request was blocked by test-account model authorization. Full typecheck is blocked by pre-existingpackages/web-shelldaemon SDK type errors on currentmain.samplingParams.max_tokensbehavior is unchanged.Linked Issues
Resolves #6734