Model behaviour issues on Windows — page 3
Issues the maintainers labelled both platform:windows and area:model.
863 issues · 285 open · 578 resolved (67%) · first seen May 31, 2025
Is this getting better or worse?
This class of problem is holding steady. 408 new reports in the last 90 days vs 341 in the 90 before — +20%. The open backlog peaked at 285 in 2026-08 and sits at 285 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
539 of these reports name the Claude Code build they were running, spanning 142 releases. Heaviest reporters:
- Claude Code v2.1.233
- Claude Code v2.1.71
- Claude Code v2.1.119
- Claude Code v2.1.237
- Claude Code v2.1.215
- Claude Code v2.1.241
- Claude Code v2.1.92
- Claude Code v2.1.251
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 37 days across 578 closures. Of the 578 closures with a recorded reason, 12% were closed as completed and 511 as not-planned or duplicate. The most recent completed fix landed in 2026-08.
Workarounds reported by the community
105 of these issues post a workaround someone says worked. The highest-engagement ones:
- [BUG] Advisor always "unavailable" with Fable 5 advisor (Opus 4.8 main) across all sessions (v2.1.198)
- Right now, my workaround is just launching a Fable 5 agent. Not ideal since seamless handoff is preferred, but it works. Does anyone know if the advisor mechanism actually saves on Fable 5 tokens, considering how differe…
Found in the thread of #73365 · resolved - Recurring error: "The model's tool call could not be parsed (retry also failed)" interrupts sessions
- Workaround: use an English or French session instead of Chinese / Japanese First-person data point supporting the non-ASCII trigger suspected here: I kept hitting while working in Chinese . Opened a new session in French…
Found in the thread of #63875 · resolved - [BUG] Can't see Claude's thinking anymore
- …problem as you. Looking through other issues, I found this workaround: Hopefully they fix it in general soon. In many cases, Claude is unusable when the thinking is not displayed. It runs for minutes, burns tokens, and…
Found in the thread of #24285 · still open - Subagent model routing is broken — all mechanisms resolve to parent model (Opus)
- …pe), which the changelog doesn't mention. The full-model-ID workaround noted above likely still applies there.
Found in the thread of #43869 · still open - Claude Code changed default model to 1M context without notifying Pro users
- …overnight. I lost hours trying to debug the issue, applying workarounds (environment variables, settings.json), and ultimately had to downgrade to a smaller context model — losing functionality I was already using and pa…
Found in the description of #62199 · still open
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 863 matches; the statistics above cover all 863.
- [BUG] False-positive Usage Policy (AUP) violation on benign FPGA/hardware engineering waveform analysis
- [BUG] Claude Code repeatedly generates invalid tool invocations
- Claude not following instructions and jumping rails.
- [Bug] Context display shows incorrect token estimates for Opus 4.7 model
- Opus 4.7 regression: weaker keyboard-language typo detection + less precise memory/MD file writes vs Opus 4.6
- [BUG] Error: Extra usage is required for 1M context · run /extra-usage to enable, or /model to switch to standard context
- [Bug] 1M context window not auto-applied on Max plan in Claude Code CLI v2.1.114 (plus: `--model` flag silently drops `[1m]` suffix)
- Model skips mandatory CLAUDE.md session-start instructions when task context is present
- [MODEL] Claude repeatedly takes unauthorized server actions and fabricates data despite 41 documented corrections
- [Feature Request] AskUserQuestion usage declining with system prompts
- [MODEL] Claude Code stopped delegating to custom subagents and loading skills — regression persists in Opus 4.6
- [BUG] `context: fork` and subagents appear to use a different (lighter) model despite model configuration
- [BUG] Claude Code Caused Production Server Outage and Data Loss
- Claude recreates existing files from scratch after session compaction — CLAUDE.md instructions ignored
- Opus 4.6 (1M): Ignores CLAUDE.md rules and user instructions in long sessions
- [Bug] Reasoning model failing consistently at night
- Claude Code refuses direct user requests to file GitHub issues about its own failures
- [BUG] Edit tool gets stuck targeting wrong file in a loop (path contamination)
- [BUG] sudden nonsense responses all over the CLI version.
- [BUG] Claude Code systematically ignores rules from memory/CLAUDE.md on every edit/write
- Claude Code destroyed gitignored files with unnecessary git clean -fd during branch creation
- [Bug] Anthropic API Error: Invalid thinking parameter in requests
- [Bug] Degraded Code Generation Accuracy and Instruction Following
- [Bug] Sonnet 4 Model Consistently Introduces Non-Standard Code Patterns
- [Bug] Opus Model Access Blocked Despite MAX 20x Plan Subscription
- [BUG] Opus 4.1 model is not using MCP servers
- Unhandled Anthropic API Response Containing Inappropriate Language
- False-positive reasoning_extraction refusal blocks both benign thinking-summary discussion and incident investigation
- frequesnt and invalid Fable 5's safeguards flag
- [Bug] Sonnet 5 deletes entire folder contents while attempting file enumeration
- [Bug] Fable 5 fallback to Opus 4.8 preserves effort level instead of mapping to capability-equivalent tier
- claude-opus-4-8: malformed tool-call wrapper (raw <invoke> XML leaks as text, tool call fails) + off-topic/hallucinated replies since ~2026-06-26
- [BUG] Claude is not available - Money or compensation!
- Opus 4.6: Poor instruction adherence and context retention in long sessions
- [MODEL] Korean output corruption — words replaced with "영역" token in long responses
- [BUG] Cross-session blindness causes three-stage failure: incomplete refactor → misdiagnosis → unauthorized revert
- Memory and instructions not effectively followed across sessions - repeated mistakes waste tokens and user time
- Claude Code agent wasted ~5 hours of user time across ~10 deployments — full incident log
- [BUG] own claude said to me when it was terrible and made me lost money, time and reach the limit improperly.
- [Bug] Repeated tool hallucinations causing workflow disruptions and token waste
- [Bug] Infinite thinking loop consuming all tokens without tool execution
- [BUG] Ignoring CLOUDE.MD instructions
- [BUG] Inconsistencies between Claude Code CLI and Claude Code in Desktop app: MCPs, Skills, async usage, and opusplan mode
- [BUG] Claude Code not respecting CLAUDE.md, unresolved and well-documented bug that has been reported for at least a month.
- Model inflates trivial tasks into complex ones to avoid execution; only acts under user escalation
- Claude claims to write memory/save plan but silently doesn't, then context resets
- [FEATURE] Raise default thinking budget / auto-escalate on diagnostic prompts
- [BUG] Claude Code generated automated claude-bridge subprocess that caused account suspension — no-code user, $367 lost
- CLAUDE.md rules not reliably enforced — model ignores its own checklist
- [BUG] my issue were not solved keep on hallucinating
- [Bug] Agent ignores user instructions and responds unpredictably
- Completion theater: Claude Code declares multi-session projects "complete" without valid verification
- [Bug] Claude ignores memory file instructions and hallucinates code behavior
- [BUG] Model retries identical failing Edit tool call multiple times without diagnosing error
- [BUG] Claude Code repeatedly uses Bash syntax (extglob) on Windows PowerShell and fails to adapt
- [BUG] No longer follows CLAUDE.MD instructions for reading procedures
- Opus 4.6 hallucinations & issues - Presents simulated data as real & ohters things.
- Add simple model configuration command and new model notifications
- context burning in Opus when switch to Sonnet is ignored
- [Bug] Unable to access Claude Sonnet model via /model command
- [Bug] Anthropic API Error: Internal server error with Claude 4.5 Sonnet
- [BUG] Changing model in a different session affects ALL sessions across different projects
- [BUG] Claude Code performed destructive mass file operations beyond requested scope
- Œ
- Missing Opus Model Option on windows Claude Code CLI
- Model Version Display in CLI Prompt
- [BUG] Stuck with Sonnet model as a Max user
- Fable 5 safeguard false-positive [cyber] blocks writing a regression test for our OWN security fix — and the recovery UI causes silent model downgrades
- [BUG] A complete CLAUDE.md rule contract governed nothing: 9 fabricated causes, stale state asserted as current, acceptance step silently skipped across a 4.5h session
- [BUG]
- [Bug] Profile bio field incorrectly triggers safety filters
- Hanja/foreign character substitution recurs in normal chat responses, not just suggestion box
- Agent asserted unverified causal explanations as fact and wrote them into permanent project records (5+ times in one session)
- [BUG] Fable 5 appended a fabricated user turn to the end of its own assistant response (JSONL-verified, ~30% of 1M context)
- Python file I/O snippets should always include encoding='utf-8' to prevent silent data corruption on Windows (follow-up to #62828)
- [BUG] Claude Code wastes paid sessions fixing self-created bugs — no accountability, no forward progress
- [BUG] /model does not display sonnet[1m] or opus[1m] despite both models being accessible via --model
- [MODEL] Claude Code safe-mode --print/resume misroutes conceptual prompt as tool-task clarification
- [MODEL] Opus 4.8 convinced me I was under a prompt injection attack .It startled me. After repeated verification, it turned out to be an illusion. It was just that the process was so realistic.
- [BUG] Sometimes responses are generated that ignore the language setting.
- [BUG] Claude Code stuck with "Usage credits required for 1M context" even after /clear and available usage limit
- [BUG] Model selection not persisting - Max subscription triggers usage credits error for 1M context
- Subagent ignores explicit instruction to skip test execution
- [BUG] Claude Code v2.1.160 forces 1M context on Windows VS Code, blocks session continuation after error
- [BUG] Why I can not use Sonnet even I have enough quota?!
- Claude Code (Opus 4.7) silently narrows user-assigned task scope using self-generated safety heuristic, even when explicit in-session directive and persistent memory forbid it; laundered through external reviewer agent
- Claude repeatedly ignores tool rejections and burns user tokens
- App Build with Terrible Results
- [BUG]
- [BUG] API error - AUP on Claude Code Opus 4.7
- [BUG] Bash tool call rendered as raw XML instead of executing
- [BUG] Token waste due to repeated failed edits that broke a live production site
- [BUG] Korean text degrades to repeated "측" tokens in long mixed-language responses
- [BUG] Claude Code Hangs and Thinks Longer on Godot Specific Projects
- claude-sonnet-4-6 /context reports 200K — same registry-stale bug already fixed for Opus 4.7 in v2.1.117
- [BUG] Claude Code — Rule Adherence Failures
- [Bug] Session termination not respecting configured conversation continuity settings
- [BUG] Claude reads rules but don't follow it.
- [BUG] Claude constructs &&-chained command with ||-tolerant loop + trailing rm -rf, swallowing failures and wiping working tree
- [Bug] Claude Code consistently delivers code with undetected logical errors before declaring tasks complete