GPT-5.3 Codex (High): benchmark results
GPT-5.3 Codex evaluated at the high reasoning-effort setting. Provider: OpenAI. Released 2026-02-05. Access: API.
Aggregate Rating 1804 ± 15, rank #132 of 2168 rated models, from 61 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ORAgentBench - Medium | 43.9 | Pass rate (%) on the 41 medium tasks: a task passes when the | 100 |
| ORAgentBench - Medium Quality | 37 | Mean normalized objective quality (0-100) on the 41 medium t | 100 |
| SCDBench (Compilation Repair) - ABI Recovery | 89.6 | ABI recovery F1 (%; micro F1 of the recovered public functio | 100 |
| SCDBench (Compilation Repair) - Compilation | 90.33 | Compilable decompilations (%; the recovered Solidity compile | 100 |
| SCDBench - ABI Recovery | 76.6 | ABI recovery F1 (%; micro F1 of the recovered public functio | 100 |
| SCDBench - Compilation | 70.17 | Compilable decompilations (%; the recovered Solidity compile | 100 |
| SCDBench - Format Completeness | 94 | Complete responses (%; parseable compiler settings, self-con | 100 |
| SlopCodeBench (Anti-Slop) | 11.2 | Strict solve rate (%): share of the 196 checkpoints whose wo | 100 |
| ORAgentBench - Hard | 20.59 | Pass rate (%) on the 34 hard tasks: a task passes when the s | 96.2 |
| ORAgentBench - Hard Feasibility | 32 | Mean feasibility (0-100) on the 34 hard tasks: share of subm | 96.2 |
| ORAgentBench - Medium Feasibility | 68 | Mean feasibility (0-100) on the 41 medium tasks: share of su | 96.2 |
| SlopCodeBench (Just Solve) | 11.2 | Strict solve rate (%): share of the 196 checkpoints whose wo | 94.1 |
Interactive version: theaggregate.ai/model?slug=gpt-5-3-codex-high · How It Works · Data refreshed daily, snapshot 2026-10-20.