GPT-5.5 Codex: benchmark results
Provider: OpenAI. Released 2026-04-23. Access: API.
Unified ELO 1959 ± 32, rank #31 of 2656 rated models, from 9 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Vals AI SkillsBench | 62.55 | Accuracy (%) | 91.2 |
| Vibe Code Bench v1.1 | 58.19 | Score (%) | 77 |
| Vals AI Vibe Code Bench | 58.19 | Accuracy (%) | 65.6 |
| Vals AI SWE-bench Verified | 76.4 | Resolved (%) | 56.9 |
| Lean AI Formalization Leaderboard | 2 | Solved Problems | 53.2 |
| ActiveVision (Agents) | 37.6 | Accuracy (%) | 50 |
| Vals AI Terminal-Bench 2.1 | 57.3 | Accuracy (%) | 44.5 |
| BioSecBench-Function | 36.5 | Endpoint pass rate (%) | 38.1 |
| VariantBench | 28.8 | Endpoint pass rate (%) | 10 |
Interactive version: theaggregate.ai/model?slug=gpt-5-5-codex · How It Works · Data refreshed daily, snapshot 2026-09-19.