Introducing GPT-5.3-Codex Feb 5, 2026
- SWE-Bench Pro (Public) 56.8
- Terminal-Bench 2.0 77.3
- OSWorld-Verified 64.7
- GDPval (wins or ties) 70.9
- SWE-Lancer IC Diamond 81.4
GPT-5.3 Codex
GPT-5.3 Codex ranks first on 2 of the 72 benchmarks with five or more models scored.
73
Benchmarks with a score
2
Ranked first, out of five or more models
9
In the top three
OpenAI Feb 2026
GPT-5.3 Codex
Closed weights Flagship
155.7
#56 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- SWE-bench Verified 74.8% #15 /32
- Terminal-Bench 2.0 78.4% #4 /43
- Toolathlon 51.9% #6 /50
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DesignArena ASCII Art | 1,147 | #48 of 77 | Official leaderboard | |
| BullshitBench V2 | 24% | #86 of 129 | Official leaderboard | |
| BullshitBench | 14.6% | #88 of 107 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-Lancer | 81.4% | #1 of 10 | Lab model card | |
| WeirdML | 77.9% | #13 of 112 | Official leaderboard | |
| METR Time Horizon | 5.8 h | #5 of 26 | Collected by Epoch AI, source not named | |
| METR Time Horizon 1.1 | 5.8 h | #5 of 26 | Official leaderboard | |
| DataPrep-Bench Finance | 59.3 | #1 of 5 | Official leaderboard | |
| ALE-Bench | 1,934 | #26 of 126 | Official leaderboard | |
| IOI (Vals) | 43.8% | #12 of 55 | Official leaderboard | |
| LiveSQLBench | 30% | #9 of 36 | Official leaderboard | |
| SlopCodeBench | 26% | #8 of 22 | Official leaderboard | |
| DataPrep-Bench Law | 76.3 | #2 of 5 | Official leaderboard | |
| OpenAI-Proof Q&A | 6.00% | #4 of 9 | Lab model card | |
| ts-bench v1 | 96% | #4 of 9 | Official leaderboard | |
| SWE-bench Verified | 74.8% | #15 of 32 | Epoch AI run | |
| Vibe Code Bench | 61.8% | #59 of 120 | Official leaderboard | |
| APEX-SWE | 41.5% | #32 of 65 | Official leaderboard | |
| SWE-Bench Pro | 56.8% | #41 of 81 | Lab model card | |
| IOI v2 (Vals) | 53.8% | #23 of 45 | Official leaderboard | |
| PostTrainBench | 17.8% | #12 of 23 | Official leaderboard | |
| SWE Atlas Refactoring | 42.4% | #9 of 17 | Official leaderboard | |
| SWE Atlas Test Writing | 39% | #12 of 22 | Official leaderboard | |
| DataPrep-Bench General | 77.6 | #3 of 5 | Official leaderboard | |
| DataPrep-Bench Medical | 35.7 | #3 of 5 | Official leaderboard | |
| DesignArena Game Dev | 1,151 | #86 of 143 | Official leaderboard | |
| SWE Atlas | 32.6% | #14 of 22 | Official leaderboard | |
| Code Arena | 1,406 | #83 of 128 | Official leaderboard | |
| DesignArena Website | 1,156 | #99 of 151 | Official leaderboard | |
| DesignArena SVG | 1,119 | #70 of 103 | Official leaderboard | |
| DesignArena Code | 1,135 | #100 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,128 | #98 of 139 | Official leaderboard | |
| DesignArena Data Visualization | 1,162 | #85 of 118 | Official leaderboard | |
| WebDev Arena | 1,370 | #76 of 105 | arena.ai, via Epoch AI | |
| Image-to-WebDev Arena | 1,444 | #44 of 55 | Official leaderboard | |
| Monorepo-Bench | 56% | #4 of 5 | Lab model card | |
| DesignArena 3D Design | 991 | #120 of 136 | Official leaderboard | |
| DesignArena Agentic Overall | 1,010 | #67 of 74 | Official leaderboard | |
| DesignArena Frontend Web App | 1,014 | #52 of 53 | Official leaderboard | |
| DesignArena React Native | 1,048 | #55 of 56 | Official leaderboard | |
| DesignArena Full Stack | 962 | #49 of 49 | Official leaderboard | |
| DataPrep-Bench Math | 22.9 | #5 of 5 | Official leaderboard | |
| DataPrep-Bench Science | 19.8 | #5 of 5 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnakeBench | 33 | #14 of 188 | Official leaderboard | |
| Terminal-Bench 2.0 | 78.4% | #4 of 43 | Official leaderboard | |
| Toolathlon | 51.9% | #6 of 50 | Official leaderboard | |
| ClawProBench | 65.5% | #9 of 49 | Official leaderboard | |
| LLM Chess | 1,053 | #24 of 123 | Official leaderboard | |
| WeaveBench | 18.4% | #5 of 9 | Official leaderboard | |
| OSWorld-Verified | 64.7% | #26 of 41 | Lab model card | |
| RuneBench | 4.29 | #50 of 76 | Official leaderboard | |
| Long-Horizon-Terminal-Bench (LHTB) | 21.5% | #24 of 25 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProfBench | 57.2% | #10 of 72 | Official leaderboard | |
| prinzbench | 52.0 | #8 of 26 | Official leaderboard | |
| QF-Bench | 60.8% | #3 of 9 | Official leaderboard | |
| APEX-1 | 65.9% | #26 of 71 | Official leaderboard | |
| Vending-Bench 2 | $5,940 | #26 of 67 | Official leaderboard | |
| GDPval | 70.9% | #8 of 10 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ArtifactArena | 1,173 | #6 of 21 | Official leaderboard | |
| MineBench | 1,564 | #34 of 75 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProtocolQA Open-Ended | 44% | #4 of 27 | Lab model card | |
| Capture the Flag (Professional) | 88% | #5 of 21 | Lab model card | |
| nocot-bench (NCRI) | 85.2 | #46 of 185 | Official leaderboard | |
| BinaryAudit | 42.1% | #5 of 20 | Official leaderboard | |
| Destructive action avoidance | 88% | #2 of 8 | Lab model card | |
| Cyber Range | 80% | #3 of 10 | Lab model card | |
| CVE-Bench | 90% | #5 of 14 | Lab model card | |
| UK AISI and CAISI simulated enterprise network range | 11.0 | #3 of 6 | Official leaderboard | |
| Tacit Knowledge and Troubleshooting | 72% | #16 of 30 | Lab model card | |
| TroubleshootingBench | 32% | #14 of 20 | Lab model card | |
| Multimodal Troubleshooting Virology | 50% | #20 of 27 | Lab model card | |
| CyScenarioBench | 0% | #15 of 17 | Lab model card | |
| Apollo Research sabotage evaluations | 0.88 | #3 of 3 | Lab model card |
Its lab's numbers put it at 155.7, number 56 of 404
The Self-Reported Index fits one score to the 6 capability results OpenAI reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 150.7 to 160.7. How we compute it.
- SWE-Lancer 81.4
- Terminal-Bench 2.0 77.3
- GDPval 70.9
- OSWorld-Verified 64.7
- SWE-Bench Pro 56.8
- OpenAI-Proof Q&A 6
Its lab's number next to a leaderboard
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| Terminal-Bench 2.0 | 77.3% Terminal-Bench 2.0 | 78.4% Leaderboard, SageAgent | −1.1 pts |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 15 of them aren't in the launch post.
GPT-5.3-Codex System Card Feb 5, 2026