Claude Code changelog

GitHub issue claims Opus 5.5 quality fell back to Opus 5 level

A single Claude Code GitHub issue on version 2.1.284 claims Opus 5.5 quality degraded to the level of Opus 5. It was labeled needs-info and has no Hacker News thread. Livenerf found no change in the same period.

Livenerf starts a 30-day test of whether Opus 5.5 gets worse after launch

Livenerf runs a frozen 78-question panel (from GPQA Diamond, MMLU-Pro, competition math and AIME) through headless Claude Code on a Max plan, 90 samples a day for 30 days starting September 24, 2026, with a pinned CLI and exact-match grading. The Hacker News launch thread drew 926 points and 394 comments.

Who is affected:

Claude Max subscribers who suspect Opus 5.5 has been quietly downgraded.

What to do:

No action needed. Watch for the first decision around October 24, 2026.

MarginLab Claude Code tracker collects an Opus 5.5 baseline, with detection paused

As of its September 29, 2026 update, MarginLab's Claude Code tracker (Opus 5.5 at high effort, from September 24) showed a 77% daily pass rate on 50 tasks, 76% over 7 days (350 tasks) and 81% over 30 days (1,200 tasks). Degradation detection is paused until a baseline exists.

Who is affected:

People watching Claude Code with Opus 5.5 for silent regressions.

What to do:

No action needed. Check the tracker again once the baseline is set.

Report: Claude Code 2.1.237 sent the 'high' effort setting as the old 'low' value

A MarsBit report on KuCoin says a developer found Claude Code 2.1.237 mapped the 'high' reasoning level to the value 10, previously used for 'low', as part of an undisclosed A/B test, and that an Anthropic engineer confirmed the issue but said it did not affect model performance.

Who is affected:

Claude Code users on version 2.1.237 who selected high effort.

What to do:

Update Claude Code and confirm the effort level with /effort.

MarginLab tracker measures a five-day Claude Code drop tied to CLI version 2.1.150

MarginLab's daily SWE-Bench-Pro subset run of Claude Code (50 tasks per day) fell from a 65% Opus 4.7 baseline to 50 to 54% from May 22 to May 26, 2026, and recovered to 66% when version 2.1.153 landed. Tool calls per task rose about 60% over the same days, and MarginLab concluded it was a harness issue, not a model regression.

Who is affected:

Claude Code users on CLI versions 2.1.150 to 2.1.152 in late May 2026.

What to do:

No action needed. Later CLI versions recovered.

Claude Code regression: malware reminder makes subagents refuse legitimate edits

A GitHub issue opened April 16, 2026 reported that Claude Code v2.1.111 still injects a malware system reminder into every file read, causing Opus 4.7 subagents to refuse code edits, after an earlier fix in v2.1.92. The thread drew 252 points and 147 comments.

Who is affected:

Claude Code users running subagents that edit code.

What to do:

Update Claude Code and report refusals with /feedback.

Anthropic confirms three changes degraded Claude Code quality

Anthropic traced user reports of worse Claude Code answers to a lower default effort (March 4 to April 7), a bug that dropped older reasoning every turn (March 26 to April 10) and a verbosity prompt change (April 16 to 20). The API was not affected, and Anthropic reset usage limits for all subscribers on April 23.

Who is affected:

Claude Code, Agent SDK and Cowork users on Sonnet 4.6, Opus 4.6 and Opus 4.7 between March and April 2026.

Action needed:

Update Claude Code to 2.1.116 or later.

AMD AI director reports Claude Code got dumber and lazier since February

The Register reported that AMD's AI group director analyzed 6,852 Claude Code sessions with 234,760 tool calls and concluded Claude Code could not be trusted for complex engineering tasks. Anthropic later attributed the period to three product changes.

Who is affected:

Teams running long, complex Claude Code sessions in February to April 2026.

What to do:

No action needed. See the April 23 postmortem.

Hacker News thread: Claude Code is being dumbed down?

A blog post argued that Claude Code 2.1.20 replaced file paths and search patterns with one-line summaries such as 'Read 3 files', hiding what the agent does. The thread drew 1,085 points and 701 comments.

Who is affected:

Claude Code users who review which files the agent reads.

What to do:

Use verbose mode if you need full file paths.

Anthropic confirms three infrastructure bugs degraded Claude responses in August and September 2025

Anthropic published a postmortem saying three separate infrastructure bugs, the first introduced on August 5, intermittently degraded Claude's response quality until early September. A load balancing change on August 29 increased affected traffic. The Hacker News thread drew 381 points and 116 comments.

Who is affected:

Users of Claude Sonnet 4, Opus 4 and 4.1, and Haiku 3.5 on the Claude API and in Claude Code between August 5 and mid-September 2025.

What to do:

No action needed. Anthropic says all three bugs are fixed.