Codex raises WebSocket limit 16MB→128MB, oversized requests now ~5x faster
charliermarsh · x · 2026-11-21
charliermarsh details Codex's transport-layer overhaul:
- Model requests stream incrementally over WebSocket; the old 16MB limit forced large turns into slow reject/retry cycles and HTTP fallback.
- The limit is now 128MB (8x), so large turns avoid HTTP fallback entirely.
- Requests that do exceed the limit are handled 5x faster.
- Overall HTTP fallback rate dropped 20%, a large number given the denominator includes all fallback causes (connection failures, timeouts, etc.).
A first-hand engineering postmortem of a real bottleneck as agent contexts grow.
Related event: Codex raises WebSocket limit to 128MB, cutting HTTP fallback ~20%(5 posts)→
More from coding & agent
- CapCut ships ChatGPT plugin for prompt-to-video with editable timeline via MCP — HeyZoyaKhan · 2026-11-21
- ENZO open-source self-hosted AI workspace unifies free-tier APIs and local models — anonymous20256 · 2026-11-21
- Codex raises WebSocket limit to 1 GB, big speed gains on large turns — charliermarsh · 2026-11-21
- Codex raises WebSocket request limit 8x to 128MB, making large turns 5x faster — charliermarsh · 2026-11-21
- Neon boosts Antigravity cache hit rate ~4%, raising quota effectiveness by at least 20% — cgarciae88 · 2026-11-21
- ArcFront launches with 4 of 5 frontier labs as customers, targeting agent evals — eptwts · 2026-11-21