ChatGPT Plus gets the Codex and Work 5-hour cap back — Pro $100/$200 stay uncapped for now.
OpenAI engineering lead Thibault Sottiaux said Plus accounts regain a five-hour limit across ChatGPT Work and Codex starting 25 Aug, after weeks with only the weekly cap.12 The stated reasons: smooth compute load so the weekly allotment can stay generous, and stop casual Plus users burning a full week in one sitting.1 After either the five-hour or weekly limit, wait for reset or buy credits; banked promotional resets still apply.1 Pro $100 and Pro $200 keep the five-hour limit off for the upcoming months; Enterprise/Edu stay on their credit system.1 HN carried the report at 112 points / 120 comments inside the window.2
npm @anthropic-ai/claude-code dist-tags this morning: stable still 2.1.231; latest/next → 2.1.246 (25 Aug 22:31 UTC).34 That line is fifteen patches past stable and includes the Bedrock double-bill and gateway cache fixes from #015, plus a large reliability batch in 2.1.246 (MCP empty-schema args, headless interrupt reporting, background session open failures, Write OOM on large overwrite).4
Plus Codex/Work: plan around a 5h rolling window again, or move Pro / credits / another harness. Claude Code on Bedrock or a gateway: stay on latest ≥2.1.239 (ideally 2.1.246); stable still lacks those fixes.
Deeper — what 2.1.246 actually fixed
- MCP: empty
{}schemas keep real arg types; interrupted tools no longer look "completed with no output".4 - Background sessions: open after sleep/deleted cwd; EACCES during npm reinstall; worktree retention no longer deletes your
.claude/worktrees/.4 - Gateways: third-party
tool_usewithoutid; resume 400s from proxy history; telemetry no longer sends gateway keys to Anthropic.4 - stable remains 2.1.231 — channel decision from #015 still stands.3