GPT-5.2 Codex: benchmark results
OpenAI's coding-specialized GPT-5.2 Codex model for software tasks. Provider: OpenAI. Released 2025-12-18. Access: API.
Aggregate Rating 1767 ± 9, rank #95 of 1656 rated models, from 189 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Context-Bench Filesystem | 93 | Rubric Score (%) | 100 |
| CyberGym-E2E (615 tasks) - PoC Generation | 30.2 | End-to-end stage S1 rate (%) on the initial 615-task release | 100 |
| MolViBench - Level 5 Create | 9.4 | Pass@1 (%) on the 64 Level 5 tasks (Create), direct generati | 100 |
| ResearchGym - Cross-Modal Retrieval | 96.6 | Best-of-3 score normalised to the task's reference (SOTA) re | 100 |
| ResearchGym - Improving Replay Buffers | 49.7 | Best-of-3 score normalised to the task's reference (SOTA) re | 100 |
| SWE-AGI (Hard) - Test-Suite Pass Rate | 91.2 | Test-suite pass rate (%, 0-100): mean over the 8 hard-tier S | 100 |
| LiveBench Code Completion | 86.96 | Score | 97.1 |
| AA IFBench | 77.62 | Accuracy (%) | 96.4 |
| 6G-Bench - AI-Native Networking and Agentic Control | 84 | Accuracy (%) on the AI-native networking and agentic control | 96.3 |
| BinaryAudit | 90.67 | Avg Success Rate (%) | 96 |
| Vals AI LiveCodeBench | 87.99 | Accuracy (%) | 94.4 |
| Tinybird AI SQL Benchmark - First-Attempt Success Rate | 100 | Questions answered with a valid query on the first attempt ( | 94 |
Interactive version: theaggregate.ai/model?slug=gpt-5-2-codex · How It Works · Data refreshed daily, snapshot 2026-10-20.