Codex's underrated feature: built-in blind A/B comparison for outputs
Angaisb_ · x · 2026-10-14
A developer highlights an underrated Codex feature: a built-in blind A/B eval UI. It shuffles outputs from two runs (e.g. two Astra Max pixel-art generations), lets you vote A/B/Tie per item, then reveals which method produced which — a handy way to compare models or sampling settings side by side.
More from coding & agent
- Multi-agent tree search: pooling discoveries can scale returns as N² — Hidenori8Tanaka · 2026-10-14
- Dev praises AI bot that proactively claimed a PR and ran code review on its own — msg · 2026-10-14
- Nous Research ships Hermes Agent v0.22.0, the 'Threshold' release — Polymarket · 2026-10-14
- Cohere Labs: scaling multi-agent teams for long-horizon software engineering — Cohere_Labs · 2026-10-14
- ResearchTwin MCP server lets agents discover researchers, papers and datasets — modelcontextprotocol · 2026-10-14
- RustGIS vs QGIS: an 18-30 MB Rust app with browser WASM build and built-in AI assistant — giswqs · 2026-10-14