Model behaviour issues on macOS — page 3
Issues the maintainers labelled both platform:macos and area:model.
3,249 issues · 731 open · 2,518 resolved (78%) · first seen May 20, 2025
Is this getting better or worse?
This class of problem is holding steady. 1183 new reports in the last 90 days vs 1426 in the 90 before — -17%. The open backlog peaked at 731 in 2026-10 and sits at 731 today.
Derived by counting each matching issue from its creation date until it closed. First-party
computation over the anthropics/claude-code tracker — the trend is not published
anywhere upstream.
Which builds report this
2,551 of these reports name the Claude Code build they were running, spanning 222 releases. Heaviest reporters:
- Claude Code v2.1.198
- Claude Code v2.1.170
- Claude Code v2.1.201
- Claude Code v2.1.207
- Claude Code v2.1.197
- Claude Code v2.1.296
- Claude Code v2.1.212
- Claude Code v2.1.199
Report volume, not a defect count: an issue can mention a build without that build having caused the failure. Versions come from the version string in the bug-report template, matched against real published tags.
Do these get fixed?
Median time to close is 36 days across 2,518 closures. Of the 2,518 closures with a recorded reason, 11% were closed as completed and 2,248 as not-planned or duplicate. The most recent completed fix landed in 2026-10.
Workarounds reported by the community
237 of these issues post a workaround someone says worked. The highest-engagement ones:
- [Bug] Anthropic API Error: Model's tool call could not be parsed (retry also failed)
- This issue makes claude code unusable... Quick workaround:
Found in the thread of #62123 · still open - [BUG] Opus 4.6 removed from Claude Desktop Code tab model picker after Opus 4.7 release
- …he option to use Opus 4.6 but not the "Coding" tab? Anyone has a workaround for the Desktop App?
Found in the thread of #49689 · resolved - Model setting changes globally across all sessions (regression)
- …ly in , so changing it in one session affects all sessions. Workaround 1 — Use project-level model override: Set the model per-project in : Project-level settings override the global setting, so each project can have its…
Found in the thread of #20745 · resolved - [BUG] Claude clobbers correct typographic marks and can't be told otherwise
- …ization code - Any scenario where Unicode precision matters Workaround: Use Unicode escape sequences (e.g., ) instead of literal Unicode characters. ------------ I've not yet tested the workaround, but I imagine it'd wor…
Found in the thread of #1599 · still open - [BUG] Claude Code defaults to 1M context on fresh session with no workaround on Pro plan
- …ars immediately without any prompt or file loaded Attempted workarounds — none resolved it: - — same error - — same error - → with Pro credentials only — same error Expected behavior: Standard context window used by defa…
Found in the description of #62063 · resolved
Excerpts are quoted from the linked issue so you can judge relevance before opening it. Threads that only ask for a workaround, or report that there is none, are excluded. These are community suggestions, not official guidance from Anthropic — read the full thread before applying one.
Matching issues 201–300
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 3,249 matches; the statistics above cover all 3,249.
- [Bug] Claude prioritizes speed over correctness - fundamentally wrong priority system for a coding agent
- [Bug] Opus 4.5 missing from Claude Code model picker menu
- [BUG] Pro account switching to sonnet[1m] started throwing API 400 error
- [Bug] Claude Code: Limited Model Selection and Frequent Context Compaction/
- [Bug] Anthropic API: Unable to Access Claude Opus with Max Subscription
- Critical Bug: Claude Code CLI is making excessive background API calls, ignoring model configuration, and console reporting inconsistencies
- [BUG] modelPicker skips an `opusplan` row as already covered, but the picker shows no Opus Plan Mode row
- [BUG] Opus 5: treats direct instructions as negotiations; injects self-referential and interpersonal content into task responses (regression vs prior models)
- [Bug] Opus4.8 slow performance across all effort levels
- [BUG] claude-opus-4-8 fabricates user messages inside thinking blocks and replies to them (~26 min self-conversation; session restart does not help, only model rollback does)
- [MODEL] Opus 4.8 fabricated a present-tense user request and persisted on an invented task context
- Regression in 2.1.150-2.1.158: malformed tool_use blocks not parseable, retry-resistant
- [Bug] Opus 4.7 & 4.8 — silent stuck-turn: stop_reason=tool_use with no tool_use block (Sonnet 4.6 unaffected)
- [Bug] Anthropic API Error: Usage Policy violation triggered on benign input
- [BUG] False positive "Usage Policy violation" errors on normal coding requests
- [BUG] Agent tool `model` parameter silently overridden by CLAUDE_CODE_SUBAGENT_MODEL — no signal in tool_result or transcript
- [Feature Request] Add opusplan[1m] preset for Sonnet 4.6 1M context support
- [BUG] Opus 4.7 (1M context) severe latency regression starting evening of Apr 24, 2026
- Opus 4.6 quality regression — confident but unverified claims increasing over time
- [BUG] Opus produces garbled multi-script output (Cyrillic/CJK/Tibetan mixed into English), triggers misleading usage policy violation
- Significant quality regression — feels like a lower-tier model is being served
- Max x20 plan blocked from using Opus 4.6 with 1M context - "Extra usage is required"
- Significant intelligence regression since 200K→1M context window upgrade
- [BUG] `API Error: Rate limit reached` Max plan ($200) cannot use model `sonnet[1m]` but can use model `opus[1m]`
- [Bug] Claude ignores explicit instructions and makes autonomous decisions instead of following specified processes
- [BUG] Selecting model with /model deletes content of settings.json
- EVER SIGNLE TASK DONE BY CLAUDE CODE HAS TO BE REDONE ADN CORRECTED BY CODEX
- [MODEL] Trailing unsolicited blocks ("Two things worth knowing...") survive CLAUDE.md rules, a Concise output style, and per-turn hook injection
- [Bug] Claude Code uses vague, made-up, or fancy wording instead of plainly describing the code
- [Bug] Anthropic API Error: Unintended safety flag on non-cybersecurity code audit request
- [Bug] Fable 5 safeguard flag fires on every message in normal conversations, not "sometimes"
- [Bug] Model silently switches from 1M to non-1M Opus mid-session, causing unrecoverable API Error
- [BUG] Model fabricates subsequent user/system turns inside one assistant message, then acts on its own fabrication
- [Bug] Overly aggressive safety guardrails blocking legitimate code security audits
- [Bug] Claude Opus 4.8 API returning increased hallucinations
- Claude Code spent a full day on a 2-minute disk reformat by refusing to stop digging
- Opus 4.7 repeatedly ignored explicit user instructions to use installed skills; caused excessive token spend over multi-hour session
- [BUG] Claude Model Missmatch
- Opus 4.6 max effort producing shallow, non-engaging responses on Pro Max plan
- Claude Code CLI Bug Report — Permission System Broken + Agent Fails to Follow Workspace Rules
- [BUG] Token burn is increasingly high since Opus 4.6
- InboxPoller delivers queued messages without context budget check, causing unrecoverable context overflow
- [BUG] - Read tool: Claude cannot interpret or see image content when reading image files
- /context command fails when thinking mode is enabled: Haiku doesn't support thinking
- m
- [BUG] CLI needs to be transparent when Opus 4.1 output is degraded deliberately
- Persistent Model Selection Across Concurrent CLI Sessions Does Not Respect Last Choice
- [Bug] Model performance degradation: Claude 5.5 Opus quality regressed to Opus 5 level
- [MODEL] Model writes "falsifiable" when "verifiable" is the obviously intended word — reproduced 4x over 4 months, project-scoped correction doesn't transfer
- Opus 5.5 ignores output-style language rule after tool chains (0.4% -> 4.2% English)
- Opus 5.5: instructed verbatim relay of a tool output over ~230 chars is never emitted as text when a tool call follows — the block is generated inside thinking (constant 274 output tokens, no text block on the wire)
- [Bug] Opus 5.5 refuses to do any cyber work despite being part of CVP.
- sonnet[1m] and fable[1m] model aliases do not apply 1M context window (only opus[1m] works)
- [Bug] Opus 4.8 Model Quality Degradation
- [Bug] Bash tool emits non-zsh-compatible syntax when SHELL=zsh on macOS
- [Bug] API Error: Usage Policy violation on simple requests
- [BUG] Everything I do is a violation
- [Regression] Opus 4.7 delivers worse quality at higher token cost vs 4.6 for production coding workloads
- [BUG] Claude Code repeatedly ignores project-level CLAUDE.md rules despite acknowledging them.
- Pro Max subscriber: quality regression cost me days of work and real money
- [Bug] Claude Code hallucinated user request during multi-turn conversation
- Agent refuses to use shell/bash under any circumstance
- Team agent teammates receive normalized model ID instead of Bedrock model ID, causing 400 errors
- [BUG] MCPSearch usage breaks Haiku
- [Bug] Opus no longer performs validation checks
- [BUG] Claude Code System Instructions Cause Claude to Lie, Fabricate Results
- Max plan models (Opus 4.8, Fable) not available in Claude Code CLI
- [Bug] Advisor tool: Fable 5 returns "unavailable" error while Opus 4.8 works
- [BUG] Remote Control: model picked at new-session start (Opus) silently reverts to default (Sonnet) within seconds
- [Bug] Fable 5 Advisor Not Available
- [BUG] Opus 4.8 fabricates user turns across 3 related sessions — one executed a real unauthorized commit+push on a fabricated confirmation (JSONL-verified)
- [Bug] Anthropic API Error: Content Policy Violation Blocks Code Scanning Analysis
- [Bug] Claude Code generates invalid Python 3.14 PEP 758 exception syntax and fails validation
- [Bug] Tool use regression in version 4.8 causing premature termination and parallel execution issues
- [Bug] Anthropic API Error: Usage credits required for 1M context window
- [Bug] Usage Policy cyber-safeguards block parallel headless Task() subagents doing localhost QA
- [BUG] Max plan: 'Usage credits required for long context requests' on standard 400k context
- [Bug] Claude model performance regression: instruction following and task completion degradation
- Opus 4.7: Excessive confirmation friction blocks autonomous workflows — regression vs Sonnet 4.6
- Write tool silently converts full-width CJK punctuation to half-width ASCII when overwriting files
- [Bug] Claude 4.7 refuses authorized security tooling work that 4.6 handles
- Brief mid-response token incoherence with self-recovery (Opus 4.6 1M, extended session)
- [Bug] Performance degradation after usage threshold suggests undisclosed rate limiting or model degradation
- [BUG] Model attempts file edits without reading file first, wasting tokens and time
- Model setting in settings.json is ignored on startup - defaults to Sonnet instead of configured Opus
- [BUG] Claude Code seems to use Sonnet even when configured explicitly to use Opus
- [BUG] Claude Systematically Violates Explicit Zero-Tolerance Policies by Inserting Placeholders Instead of Failing Explicitly
- switch to opus causes usage policy violation error
- [BUG] Claude CLI fails to escape file paths with spaces in bash commands
- [BUG] Answers have become gibberish and semantically incoherent
- [Bug] Anthropic API Error: Usage credits required despite extended plan validity
- [Bug] Fable 5 model missing from model picker after API incident
- [Bug] Opus model experiencing high latency and frequent API errors in Claude Code
- [Bug] Non-ASCII characters appearing unexpectedly in Claude Code output
- [BUG] Claude Code commits without being asked, repeatedly ignoring CLAUDE.md rule
- [Bug] Anthropic API Error: Invalid or inaccessible model claude-fable-5
- [Bug] Overly aggressive safety filtering blocks legitimate chemistry and biology queries
- False-positive 'cyber-related safeguards' block on hard-SF fiction Q&A — fires on tool-result continuation, re-fires on benign retry
- [BUG] Long-running session degradation since v2.1.139, compounded by silent fast-mode model swap in v2.1.142 — no changelog, breaks pinned-model workflows, quantified waste
- [BUG] Intermittent Korean syllable corruption in AskUserQuestion — wrong valid characters generated (not U+FFFD)