Model behaviour issues on Windows — page 2
Issues the maintainers labelled both platform:windows and area:model.
860 issues · 283 open · 577 resolved (67%) · first seen May 31, 2025
Is this getting better or worse?
This class of problem is holding steady. 405 new reports in the last 90 days vs 342 in the 90 before — +18%. The open backlog peaked at 283 in 2026-08 and sits at 283 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
536 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 v2.1.158
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 577 closures. Of the 577 closures with a recorded reason, 12% were closed as completed and 510 as not-planned or duplicate. The most recent completed fix landed in 2026-08.
Workarounds reported by the community
104 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 101–200
Ranked by community engagement (reactions weighted over comments). Listing the top 300 of 860 matches; the statistics above cover all 860.
- [BUG] Unauthorised launch of extremely inefficient subagents that wasted tokens - I want a refund
- Tool calls rendered as raw text and fail to parse with claude-opus-4-8 in VS Code extension
- Model compliance regression: auto-loaded .claude/rules/ procedures skipped on session start
- [Bug] Claude prioritizes apologetic acknowledgment over actually fixing the reported error
- [Feature] Completion summaries should distinguish verified facts from inferred claims
- [BUG] 500 No available Claude accounts support the requested model: claude-sonnet-4-5
- Safeguard model-switch is sticky — session never returns to the selected model after a flagged turn
- Malformed tool-call emission: literal "court" replaces <function_calls> opener, antml: namespace stripped, block stored as text (session hangs)
- [BUG] /model picker shows Fable 5 as disabled while --model claude-fable-5 works on the same account
- [MODEL] claude-opus-4-8 hallucinated destructive OS commands (rm -rf / del /s /q) as prompt-injection warnings — confirmed clean via JSONL
- [BUG] Max plan: Opus 4.8 only shows 256k context, 1M option missing in model picker
- [BUG] Model fallback downgrade is disruptive and doesn't stay pinned (Fable→Opus mid-session)
- [Bug]: Fable 5 flags benign greeting ("hi") due to security-related filenames in system-injected git context
- [BUG] False positive safety block: triggered without apparent reason during game dev session
- Model switched to Opus without consent or disclosure — five process failures, seven cost amplifiers, $1,050 overcharge on May 5-7
- [BUG] Text-only status messages mid-task silently end the turn and hand control back to the user
- [BUG] System instructions enforces push workflow by default, contradicting user control principle
- [BUG] API Error (model ref): 400 The provided model identifier is invalid.. Try --model to switch to us.anthropic.claude-sonnet-4-5-20250929-v1:0.
- [BUG] New version burns through all tokens in minutes, doesn't allow Opus 4.6 use, doesn't allow downgrade.
- Own code, market as Malware and refuses to make edits, after an hour I still worked what is happening?
- [Bug] File Read Encoding Corruption Triggers False Positive Usage Policy Block
- [BUG] The API content filter is blocking legitimate pastoral and theological writing
- [BUG] Does not Follow Claude.MD
- [Bug] Claude generates incorrect code/SQL and reports it as complete without validation
- [Bug] Claude Code reports task completion without actually executing operations
- [BUG] VS Code Extension ignores model settings
- [BUG] Claude Code hangs on in loading model when press /model
- Claude Code: Missing Opus Model Option Despite Max Plan Access
- [BUG] Context corruption ? fabricated file writes, injected tool-result text, and confabulated user turns leading to an attempted destructive git action
- [BUG] Sonnet 4.6 blocked by forced 1M context mode requiring additional credits on Pro plan
- Opus 4.7 1M context model option missing from /model picker after 2.1.133 upgrade
- [Bug] Anthropic API Error: Usage Policy false positive issue blocking operations
- [BUG] Claude Code for VS Code selects the wrong model, mismatches whats displayed in UI.
- Claude, the Liar who knows that he is and decides to go on.
- Bug: Claude Code skips mandatory session-start files to optimize for speed over correctness
- [BUG] [MODEL] claude-opus-4-5-20251101 - 77K apparent working context??
- [Bug] Model Performance Degradation Across Sessions
- [Bug] Sonnet 4.5: Decreased Code Generation Quality and Completeness
- [BUG] Claude code returns 504 response from Claude Code
- [Bug] Stability Issues in v1.0.103: Model Performance and Configuration Failures
- [Bug] Fable model inaccessible despite included in max plan and available usage credits
- [BUG] What is happening?? Claude code went from hero to zero this week!.
- Output enters an infinite loop repeating the token "court"
- [BUG] Fable 5 thinking blocks render as empty unclickable stubs in VS Code extension (v2.1.170)
- Error: 'usage credits required for 1M context' when session grows long on non-1M model
- [BUG] Pro plan — 1M context error blocks all requests
- [BUG] Thinking text empty / "Thought for Xs" not expandable on Opus 4.7+ in VS Code extension — fallback thinking-config branch omits `display`
- [Bug] Safety classifier blocks routine code analysis for paid users (started 2026-05-23)
- [BUG] False positive - legitimate medical-education content flagged as unsafe
- [BUG] AUP repeatedly throwing false positives - live issue ongoing - hundreds of similar reports
- [BUG] Model repeatedly begins implementation without explicit user approval
- [BUG] You made Opus dumb again and I'm officially moving to codex.
- [Bug] v2.1.121: Opus 4.7 (1M context) max-effort window silently self-downgrades to Sonnet 4.6 mid-session AND /compact does not fire at context limit
- Opus 4.7 [1M] at max effort + thinking ON: Severe task quality regression, plus silent downgrade of effort setting mid-session
- False positive AUP block during legitimate CI/CD VM provisioning session (claude-opus-4-7, 1M context)
- [BUG] claude code is broken, unable to use in vscode
- Mid-session CLAUDE.md edits don't influence behavior as strongly as startup-loaded instructions
- [MODEL] Claude Code (Opus 4.6) — Systematic Rule Violation and User Authority Override
- [BUG] Opus 4.6 at 1m context actually is Opus 4.5
- Claude Code deflects blame and offers unsolicited alternatives instead of acknowledging failures and doing what's asked
- [Bug] Multi-step tasks lack per-step verification — errors compound silently
- [BUG] switch model itself
- [BUG] Model reasoning doesn't appear to be triggered despite Thinking/Verbose both being enabled
- [BUG] Going out of my mind with Claude Code cheating.
- [Bug] Model selection defaults to Opus 4.1 when Opus 4.5 is selected
- [Bug] Falsely Claiming UI Mockups as Production-Ready Code
- [Bug] Fable 5 unavailable despite unused allowance on Max plan
- Claude fabricates task requirements when tool execution fails silently
- Malformed tool-call emission (dropped antml: namespace + spurious "court" token) under multibyte-dense context — Opus 4.8 / Claude Code
- [BUG] Claude Code long-running session invents tool results, fake user turns, and follow-up actions without user input
- [All models] Claude ignores safety-critical blocking questions and proceeds without explicit answers
- [BUG] "Usage credits required for 1M context" error when context is set to Medium on Sonnet 4.6
- [BUG] Claude Code executed destructive command (adb shell pm clear) against explicit user instruction, causing permanent data loss
- [MODEL] False-positive cyber-safety block on benign Minecraft mod documentation task
- [BUG] API Error: Extra usage is required for 1M context · run /extra-usage to enable, or /model to switch to standard context
- Opus 4.7 agent-mode Korean output regression — in-vivo self-diagnosis from the model (related to #54339)
- [Bug] Model exhibits inconsistent reasoning, instruction non-compliance, and context degradation in extended sessions
- [Bug] Claude Code 4.7 High Token Usage and Context Limitations
- Usage limits + forced T&C changes + model removal = no real user choice
- [BUG] Opus 4.7: In-session corrections not retained — structurally unreliable for autonomous operation
- Model fails to update all version locations despite explicit checklist in CLAUDE.md
- Model fabricated data in analysis script and presented it as findings
- [Bug] Claude model exhibits persistent reasoning bias and fails to accept contradicting information
- Model makes no distinction between inferred claims and verified claims — both produce equal confidence
- Claude defaults to bash/Unix syntax on Windows instead of PowerShell
- [Bug] Claude Code not following skill instructions
- [BUG] Claude ignores its own memory instructions repeatedly — behavioral rule enforcement failure
- [BUG] Plan mode violated after ultraplan approval, model edits source files instead of plan file
- [Bug] Performance degradation during night hours (EU region)
- [Bug] Claude ignores non-fatal tool output — warnings and stderr treated as success
- [Bug] Verification queries are tautological — cannot return failure results
- [BUG] VS Codium extension ignores model configuration and defaults to Opus 4.1
- [BUG] /model command shows Opus as "Default" instead of "Opus"
- [BUG] Haiku always runs upon any client init
- [Bug] Anthropic API: Opus Model Functionality Discrepancy Between CLI and Web Interface
- Fable 5 safeguards flag first-party defensive security audit, falls back to Opus 4.8
- [BUG] Opus 4.8 frequently enters long "thinking" on simple tasks without taking action — suspected regression
- [BUG] Spoofed/fabricated tool results injected into the session[Opus 4.8]
- [BUG] Claude Code repeatedly fails to follow CLAUDE.md session branch + push discipline
- Sub-agents declare work as PASS without performing actual verification against reference