Claude Code v2.1.174: Wheel Scroll Acceleration Toggle, /model Fixes, GovCloud Support, and VSCode Usage Attribution

Claude Code v2.1.174 is out with a mix of usability fixes and platform-specific improvements. Several long-standing bugs around the /model picker, background sessions, and Bedrock GovCloud are now resolved. There's also a new setting for mouse-wheel behavior and expanded attribution data in the VSCode extension.
New Setting: Mouse-Wheel Scroll Acceleration Toggle
Added wheelScrollAccelerationEnabled setting to disable mouse-wheel scroll acceleration in fullscreen mode. Developers running in fullscreen terminals can now set this to false for consistent scroll behavior.
/model Picker Fixes
The /model picker no longer hides the model family that Default resolves to. Specifically:
- Opus now appears as its own row on Max, Team Premium, and Enterprise plans.
- Sonnet now appears on Pro and Team plans.
- Opus appears on pay-as-you-go API accounts.
Also fixed: the /model picker was showing a hardcoded Sonnet version label when ANTHROPIC_DEFAULT_SONNET_MODEL pinned a different Sonnet. That label now reflects the actual pinned model.
Fixed: "Fable 5" Banner for Enterprise Accounts
The banner "Fable 5 is now consuming usage credits" incorrectly showed for enterprise accounts with usage-based billing. That's now suppressed for those accounts.
Bedrock GovCloud Fix
Bedrock GovCloud regions (us-gov-*) were deriving the wrong inference profile prefix (global instead of us-gov), causing 400 errors on derived model IDs. This is fixed — the prefix now correctly uses us-gov.
Background Session Env Inheritance
Fixed background sessions inheriting another session's ANTHROPIC_* provider env (gateway URL, custom headers, /model aliases) from the shell that started the background daemon. Each session now correctly uses its own environment.
Exit Pause Fixed
A 1-2 second pause when exiting Claude Code shortly after a shell command was interrupted or killed on macOS and Linux has been resolved.
Git Commit Attribution
Fixed git commit co-author attribution showing an incorrect model name for some models.
/advisor Dialog and allowlist
The /advisor dialog no longer pre-selects a saved advisor model that is blocked by the availableModels allowlist.
Skill Hot-Reload Optimization
Skill hot-reload was re-sending the entire skill listing when a single skill changed. Now only the changed skill is re-announced, reducing unnecessary overhead.
Workflow agent() Attribution Headers
Fixed Workflow tool agent() subagents — they were missing per-agent attribution headers. Now each subagent sends proper attribution.
[VSCode] Usage Attribution Dialog
The Account & usage dialog (/usage) now shows usage attribution breakdowns over the last 24 hours or 7 days, including:
- Cache misses
- Long context
- Subagents
- Per-skill, per-agent, per-plugin, and per-MCP breakdowns
Pre-warmed Background Workers
Fixed pre-warmed background workers failing with "Could not resolve authentication method" when claimed after sitting idle.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Text AI Watermarks Will Always Be Trivial to Remove
EU AI Act requires text watermarks by 2026, but they'll always be removable. Here's why and how SynthID works.

Google's Nano Banana 2 AI Image Model: Features and Availability
Google DeepMind released Nano Banana 2, an image generation model combining Nano Banana Pro's advanced features with Gemini Flash's speed. It offers subject consistency for up to five characters, supports resolutions from 512px to 4K, and is rolling out across Google products.

Analysis: Anthropic's actual compute costs for Claude Code users are far lower than reported $5k figure
A recent article analyzes the claim that Anthropic's $200/month Claude Code Max plan consumes $5,000 in compute, finding that actual inference costs are roughly 10% of API prices when comparing to competitive open-weight models on OpenRouter.

LLM Spatial Reasoning Tested: Sokoban Benchmark Shows ChatGPT, Qwen3.7-max, Gemini 3.5-thinking Lead
A custom Sokoban benchmark tested zero-shot spatial reasoning in LLMs with strict formatting. Only ChatGPT, Qwen3.7-max, and Gemini 3.5-thinking passed. Models like Gemini 3.5-flash and Qwen3.7-plus failed due to illegal moves or deadlocks.