Model behaviour issues on VS Code — page 2
Issues the maintainers labelled both platform:vscode and area:model.
463 issues · 168 open · 295 resolved (64%) · first seen Feb 26, 2026
Is this getting better or worse?
This class of problem is still growing. 286 new reports in the last 90 days vs 170 in the 90 before — +68%. The open backlog peaked at 168 in 2026-08 and sits at 168 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
355 of these reports name the Claude Code build they were running, spanning 120 releases. Heaviest reporters:
- Claude Code v2.1.170
- Claude Code v2.1.198
- Claude Code v2.1.207
- Claude Code v2.1.210
- Claude Code v2.1.197
- Claude Code v2.1.158
- Claude Code v2.1.239
- Claude Code v2.1.235
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 35 days across 295 closures. Of the 295 closures with a recorded reason, 10% were closed as completed and 265 as not-planned or duplicate. The most recent completed fix landed in 2026-08.
Workarounds reported by the community
37 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 - Opus 4.7 thinking summaries not rendered (VSCode extension 2.1.112)
- …ith a capability table), but the suffix variant is missing: Workaround None found that works — the filter strips 4.6 at a later stage, so a byte-level rename of → (length-preserving) has no effect. Users on 4.7[1m] lose…
Found in the description of #49902 · 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 - [BUG] Server-side experiment (`x-cc-atis`) silently removed Opus 4.8 thinking summaries, and the CLI silently self-updated even with `autoUpdates: false`. No notice, no opt-in, settings silently overridden.
- …de gap, posted the analysis ( and built and shared a public workaround so other affected users could get summaries back: a launcher that injects , the one lever that is not interactivity-gated, plus an extension patch an…
Found in the description of #75607 · still open - [MODEL] Opus 4.8 starts hallucinating results before parallel tasks finish
- …e upstream fix is unknown) 1. is the only durable user-side workaround for both hazards, per longitudinal data in [ 62123 comment 4569840435]( (4.6 clean → 4.7 onset → 4.8 worst, including 1M context). @timmywheels named…
Found in the thread of #63884 · 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 463 matches; the statistics above cover all 463.
- [BUG] Agent ignored own persisted memories repeatedly + expanded scope unilaterally
- Claude claims to write memory/save plan but silently doesn't, then context resets
- CLAUDE.md rules not reliably enforced — model ignores its own checklist
- [BUG] my issue were not solved keep on hallucinating
- [BUG] VSCode extension fails on first query after restart with "model may not exist" error
- [Bug] Agent self-approves and executes write operations without awaiting user confirmation
- [Bug] Anthropic API Error: Response exceeds 32000 output token maximum
- [BUG] Model retries identical failing Edit tool call multiple times without diagnosing error
- Fable 5 safeguard false-positive [cyber] blocks writing a regression test for our OWN security fix — and the recovery UI causes silent model downgrades
- Hanja/foreign character substitution recurs in normal chat responses, not just suggestion box
- [Bug] Safeguard falsely triggered during game design discussion, model downgrade irreversible
- [MODEL] Claude ignores CLAUDE.md read/write mode rules and modifies files without explicit permission
- Bug: Tool call intermittently emitted in a malformed envelope ("court" + bare <invoke>), so it never executes
- [Bug] Fable 5 cybersecurity classifier false positive on legitimate infrastructure administration tasks
- Fable 5 Flagged work for cybersecurity or biological topic
- [Bug] Claude Code violates architectural and testing rules under task pressure; inlines SQL, skips HTTP/JWT tests, ignores migration init patterns
- [BUG] [BUG] Claude Code VS Code Extension v2.1.161 Forces 1M Context on Pro Plan Without Request — Blocks All Usage on macOS
- [BUG] Claude Code v2.1.160 forces 1M context on Windows VS Code, blocks session continuation after error
- [BUG] Claude code increasingly disregards memory items
- Duplicate: VS Code model picker shows stale model name
- VS Code: model picker collapsed row shows stale model name (Sonnet 4.6) when Default (Opus 4.7) is selected
- [BUG] Claude Code Hangs and Thinks Longer on Godot Specific Projects
- [MODEL] Sonnet 4.6 DOES NOT FOLLOW GITIGNORE AND USES -F BY DEFAULT
- After compaction and subagent return, Main agent set's itself from Sonnet 4.6 to Opus 4.7 without user interaction
- [Bug] Claude Code performance degradation on claude-opus-4-1 with high effort settings
- Claude collapsed a staged-rollout approval into a batch execution, bypassing the verification gate the user explicitly asked for
- [Bug] Model generates identical answer options across multiple test questions
- [Bug] Claude ignores custom instructions from memory and referenced files
- [BUG]
- Claude Code ignored MEMORY.md warning and edited wrong repo for entire session
- [BUG] Claude Code builds CAPTCHA solver and tests against live system without user consent
- [BUG] Claude Code pushed broken code to production 4 times without verifying fixes, breaking live SaaS application
- [BUG] Data loss caused by Claude Code rewriting a file before extracting content
- [BUG] Bug report: assistant fabricated a user turn and system prompts inside its own response, then executed them
- Opus 4.8 (direct API): thinking text empty (signature-only blocks) despite --thinking-display summarized — started 2026-07-19 ~13:07 UTC; Fable 5 unaffected
- [MODEL] Sustained Korean-language sessions degrade into ill-formed Hangul or drift fully to English — recurring cross-model pattern, repeatedly auto-closed as stale
- Claude Code fabricated a factual claim from incomplete tool output, ignored contradicting evidence, and ran unauthorized state-changing commands
- Recurring 529 Overloaded bursts (322 events, request IDs included) + session-terminating malformed tool-call output corruption (23+ forced restarts)
- Fable 5 safety classifier force-downgrades the whole session on defensive-security work (VSCode extension, no reachable feedback route)
- Fable 5 safeguards false-positive on routine git/diagram work forces model switch; advertised /feedback (and /bug) commands don't exist
- [Bug] Fable 5 model safeguards incorrectly flagging legitimate security research context
- [BUG] Claude proceeds with unrequested actions after milestone completion without confirmation
- [BUG] User billed in tokens for the assistant's own mistakes — refund request and product fix
- [BUG] Suspicious phishing-style "API Error" rendered as assistant response
- [Bug] Agent enters infinite loop consuming tokens without activity on HTML content updates
- [Bug] Claude generates false claims about file contents without verification
- [MODEL] Japanese output corruption: model emits visually-similar but wrong kanji (Opus 4.7)
- Model loses context and misreads questions in long sessions
- [Bug] Claude API: Model not adhering to system instructions and tool usage protocols
- [BUG] /model default resolves to Opus 4.7 but UI picker shows Default = Sonnet 4.6
- [BUG]Multiple regressions: context comprehension, instruction following, and unsolicited file changes degrading across recent versions
- Claude repeatedly prompts user to stop working / defer to next session
- Claude Code lacks design completeness — implements features without full lifecycle (create without delete cleanup, storage without retrieval)
- Claude Opus 4.6 - 30+ incidents of false reporting, ignoring user instructions in single session (2026-04-09)
- [Bug] Claude ignores CLAUDE.md instruction to write plans to `plans/` directory
- [Feature Request] Implement uncertainty quantification and validation checkpoints before executing costly operations
- [BUG] [VS CODE EXT] Claude says he's going to do something and then doesn't
- [Bug] Sonnet 4.6 model regression: internal details leakage before failure
- [BUG] stuck in a deliberation loop, output hallucinations and <br>
- [BUG] Agent produces cascading failures on simple feature — 9 bugs from a login screen
- [Bug] Unescaped apostrophes in string literals break UI code generation
- Claude Code ignores its own memory/MEMORY.md rules and applies them inconsistently
- Claude Code reads files matching a user CLAUDE.md 'never read' instruction
- [Bug] Anthropic API Error: Model refuses continuation with AUP message in valid session
- [Bug] Safety Check Triggered During NPU Access Verification
- [BUG] Assistant responds to a nonexistent user message (fabricated premise) — opus-4-8, 1M mode, ~272k ctx; transcript shows no user turn / no compaction / no injection
- [Bug] Fable 5 model unsupported for design and organization tasks
- Model behavior: agent substitutes easier implementations for explicit specs, makes unverified 'done' claims, games user-defined enforcement hooks (7-day pattern)
- Sessions rerouted from pinned claude-fable-5[1m] to Opus 4.8
- Mid-session model switch not applied — UI shows the new model, session keeps running the previous one (+ silent budget loss)
- [BUG] Tool calls repeatedly emitted with bare <invoke> tags (missing antml: namespace), causing parse failures
- [Bug] Unexpected model downgrade from Claude Fable 5 to Claude Opus 4.8
- opus 4.8 xhigh acting very dumb on toy reinforcement learning
- Systemic discipline failures in a high-stakes agricultural project: unauthorized edits, fabricated completion claims, ephemeral fixes to generated files, and safety-corrupting output that passed every test (~115 documented incidents)
- [BUG] Claude offers/accepts pasting secrets (API keys, tokens, passwords) directly into chat with no upfront warning
- [BUG] Rude and pressuring behavior from Claude
- [Bug] Claude Code auto-commits all uncommitted changes instead of only modified files
- [Bug] Incorrect Italian verb conjugation of English loanwords with geminated consonants
- [Bug] Local development login flagged as offensive content by content policy
- [Bug] Model defaulting to Opus when fable configuration intended
- [Bug] Agent commits changes prematurely without waiting for user revision
- [Bug] Safety filters incorrectly flagging valid code as unsafe
- Long-session time/state drift: assistant produces wrap-up artifacts with stale 'today/tomorrow' phrasing after date-rollover system-reminder
- I don't have any bug report content to generate a GitHub issue title from. Please provide the bug report or issue description, and I'll create an appropriate technical title for you.
- [MODEL]
- [Bug] Excessive verbosity in unprompted detailed responses
- [Bug] Claude Code not respecting user-defined rules and constraints inconsistently
- [BUG] "Usage credits required for 1M context" error persists after plan recharge on VS Code extension
- [BUG] Gender Bias
- [MODEL]
- [BUG] Opus 4.7 cache hit rate collapse after May 27 incident — Messages 1.1k→88.9k in 9 minutes, $630/session
- [BUG]
- [BUG] activation of 1M context in VSCode
- [BUG] False positive safety filter block for Polish-language security audit question in own PHP CRM project (VSCode extension)
- Claude fails to execute session-start protocol defined in CLAUDE.md
- [BUG] Model suffix [1m] in settings.json ignored — context window capped at 200k instead of 1M
- [Feature Request] Improve proactive suggestion of alternative methods and best practices
- [BUG] Constant lying- when WILL IT END!!??
- [BUG] Opus 4.7 standard unavailable in VS Code extension
- [Bug] Claude ignores explicit instructions to avoid shortcuts despite custom hooks and memory configuration