GPT-5.1 Codex: benchmark results
OpenAI's Codex-specialized GPT-5.1 model for coding and software-agent workloads. Provider: OpenAI. Released 2025-11-13. Access: API.
Aggregate Rating 1721 ± 10, rank #159 of 1656 rated models, from 172 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FeatureBench | 12.5 | Resolved rate (%): share of tasks whose patched repository p | 100 |
| PrepBench | 54.9 | End-to-End Prep-Code Acc. (self-reported) | 100 |
| PrepBench - Code-to-Workflow Translation | 67.7 | Translation accuracy (%): given the ground-truth preparation | 100 |
| PrepBench - GUI Workflow | 34.6 | End-to-end GUI workflow accuracy (%): from the original ambi | 100 |
| PrepBench - Interactive Disambiguation | 51.4 | Disambiguation F1 (0-100) of the agent's clarification quest | 100 |
| PrepBench - Prep-Code Generation | 85.3 | Prep-code accuracy (%) from the disambiguated request with a | 100 |
| SWE-Lancer | 66.3 | Score (%) | 100 |
| Sonar LLM Leaderboard - Java - Bug Density | 0.52 | Bugs per 1,000 lines of code (lower is better) | 98.6 |
| AA AIME 2025 | 95.67 | Accuracy (%) | 97.2 |
| AI Chess Leaderboard (Reasoning) | 1743 | Elo | 97.1 |
| AA LiveCodeBench | 84.87 | Pass@1 (%) | 95.9 |
| AA MMLU-Pro | 86.01 | Accuracy (%) | 94.5 |
Interactive version: theaggregate.ai/model?slug=gpt-5-1-codex · How It Works · Data refreshed daily, snapshot 2026-10-20.