v4.1.4 — Fable 5 & Sonnet 5, Real 1M Context Window, 30× Faster Statistics

Highlights

🆕 Fable 5 & Sonnet 5 model support

• claude-fable-5 — 5th-gen flagship (Mythos-class), 1M context
• claude-sonnet-5 — most agentic Sonnet, new tokenizer
• Full pricing, display names, and model selector UI cards

🪟 Real 1M context window (fixes "indicator stuck at 16-18%")

• 1M-capable models are spawned with a [1m] model suffix + process-level CLAUDE_CODE_AUTO_COMPACT_WINDOW env — unlocks the CLI's 200K default ceiling without upgrading the CLI and without touching your global settings
• New setting claudeCodeChatUI.contextWindowTokens (default 400,000, range 100K–1M); denominator auto-clamps to each model's real window (Haiku stays 200K)
• compact_boundary messages are now recognized: the chat shows a "🗜️ auto-compacted" divider and the context indicator correctly falls back after compaction
• Restored history sessions no longer show stale values; context indicator driven by a backend single source of truth

⚡ Statistics panel: 47.5s → ~1.5s cold start

• Worker-thread parse pool with size-based LPT scheduling + "usage":{ prefilter
• Disk cache in globalStorage: VS Code reload warm start < 1s
• Deterministic two-phase dedup — tab counts are now consistent across runs, zero rescan on tab switch

🛡️ Opus 4.8 tool-call glitch auto-retry

• Detects the known Opus 4.8 regression where tool calls are emitted as raw text (stray "court" token) and automatically resumes with a corrective prompt (max 2 retries)

🔧 Other fixes

• Subagent (sidechain) usage no longer distorts the main context indicator
• Model pricing prefix-match fallback for date-suffixed/unknown model IDs
• New mechanism reference: docs/md/Claude-CLI-Context-Mechanism.md

Install

Download claude-code-chatui-4.1.4.vsix below, then in VS Code: Ctrl+Shift+P → Extensions: Install from VSIX

Or via CLI:

code --install-extension claude-code-chatui-4.1.4.vsix

Full changelog: CHANGELOG.md