Introducing GPT-6.1 Terra (OpenAI) 26 Oct 2026
GPT-6.1 Terra
GPT-6.1 Terra ranks first on 3 of the 70 benchmarks it has scores on.
OpenAI Released 26 Oct 2026
76
Benchmarks with a score
3
Ranked first, out of five or more models
14
In the top three
OpenAI Oct 2026
GPT-6.1 Terra
Closed weights
168.1
#3 of 271
Epoch Capabilities Index
- Agents' Last Exam 41.8% #5 /12
- MultiChallenge 74.2% #7 /35
- HealthBench Professional 62.7% #9 /25
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 | |
|---|---|---|---|---|
| MultiChallenge | 74.2% | #7 of 35 | Official leaderboard | |
| SimpleQA Verified | 57.3% | #19 of 84 | Epoch AI |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FrontierMath Tiers 1-3 v2 | 90.9% | #4 of 89 | Epoch AI | |
| GPQA Diamond | 94.3% | #10 of 219 | Epoch AI | |
| FrontierMath Tier 4 v2 | 92.7% | #4 of 71 | Epoch AI | |
| OTIS Mock AIME 2024-2025 | 100% | #13 of 202 | Epoch AI | |
| CritPt | 28% | #15 of 160 | Epoch AI | |
| LatchBio Capabilities v1.0 | 43.4% | #12 of 27 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ALE-Bench | 2,609 | #3 of 121 | Epoch AI | |
| Code Arena | 1,673 | #13 of 124 | Official leaderboard | |
| Brokk Power Ranking | 29.6 | #2 of 19 | Official leaderboard | |
| OpenDesign Arena | 90.3 | #2 of 19 | Official leaderboard | |
| WeirdML v3 | 28.1% | #4 of 25 | Official leaderboard | |
| Vibe Code Bench | 84% | #20 of 115 | Official leaderboard | |
| FrontierCS 2 | 39.6% | #5 of 16 | Official leaderboard | |
| Internal Research Debugging Evaluation | 76.4% | #2 of 6 | Model card | |
| Next.js Agent Evals | 97% | #5 of 14 | Official leaderboard | |
| KernelGen 1P | 45.3% | #2 of 4 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| EBR-bench | 78.1% | #1 of 28 | Epoch AI | |
| Mystery Game Puzzles | 74% | #3 of 80 | Epoch AI | |
| Agent Arena | 0.11 | #8 of 53 | Official leaderboard | |
| Terminal-Bench 4.0 | 50.3% | #10 of 23 | Model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PolicyBench | 93.1% | #3 of 50 | Official leaderboard | |
| APEX-Agents | 68.3% | #6 of 56 | Epoch AI | |
| Excel Modeling Benchmark v1.1 | 73.3% | #7 of 37 | Official leaderboard | |
| Vending-Bench 2 | $9,107 | #13 of 66 | Official leaderboard | |
| Vals Index v2.2 | 60.5% | #11 of 32 | Official leaderboard | |
| HealthBench Professional | 62.7% | #9 of 25 | Model card | |
| Agents' Last Exam | 41.8% | #5 of 12 | Official leaderboard | |
| Legal Research Bench | 26% | #59 of 88 | Official leaderboard | |
| Tax Agent Bench | 17.1% | #55 of 79 | Official leaderboard | |
| HealthBench Consensus | 95.1% | #12 of 17 | Model card | |
| Vals Finance Index | 53.3% | #26 of 35 | Official leaderboard | |
| HealthBench Hard | 31.4% | #16 of 21 | Model card | |
| Vals Finance Agent v2.1 | 56% | #29 of 35 | Official leaderboard | |
| HealthBench | 55.1% | #25 of 30 | Model card | |
| Vals Legal Index | 58.8% | #30 of 33 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 1.67% | #33 of 33 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| VoxelBench | 2,295 | #6 of 56 | Official leaderboard | |
| Furniture Assembly Benchmark | 79.2% | #4 of 35 | Epoch AI | |
| ParseBench | 73.6% | #5 of 42 | Official leaderboard | |
| Blueprint-Bench 2 | 42.1% | #5 of 35 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SpatialBench-Long | 40.5% | #2 of 27 | Official leaderboard | |
| Broken Search Tool | 1.83% | #1 of 7 | Model card | |
| ExploitBench | 97.1% | #3 of 19 | Model card | |
| Coronavirus-ACE2 Cell-Entry Screen | 0.47 | #1 of 5 | Model card | |
| ExploitBench - Internal Port | 24.5% | #1 of 4 | Model card | |
| SEC-bench Pro | 83.3% | #3 of 10 | Model card | |
| Blocker Deception | 8.47% | #1 of 3 | Model card | |
| FORTRESS | 18.2% | #23 of 66 | Official leaderboard | |
| Phage-plasmid Co-evolution | 10.8 | #2 of 5 | Model card | |
| SHP2 Protein Function Prediction | 0.37 | #2 of 5 | Model card | |
| TxBench-Preclinical-Pharmacology | 55.7% | #12 of 28 | Official leaderboard | |
| TxBench-Antibody-Discovery | 40.7% | #13 of 28 | Official leaderboard | |
| BioSecBench-Function | 39% | #13 of 27 | Official leaderboard | |
| EpiBench | 23.3% | #13 of 27 | Official leaderboard | |
| MetagenomicsBench | 38.7% | #13 of 26 | Official leaderboard | |
| Multimodal Troubleshooting Virology | 52.5% | #13 of 26 | Model card | |
| MentalHealthBench | 55.5% | #2 of 4 | Model card | |
| Sandbox Bench | 22.7% | #2 of 4 | Model card | |
| scBench-Long | 18.8% | #14 of 27 | Official leaderboard | |
| BioSecBench-Refusal | 34.2% | #12 of 22 | Official leaderboard | |
| VariantBench | 34.8% | #15 of 27 | Official leaderboard | |
| OncologyBench | 16% | #18 of 30 | Official leaderboard | |
| CoT-Control | 14.8% | #5 of 8 | Model card | |
| AAV Capsid Packaging Prediction (OpenAI) | 0.53 | #4 of 6 | Model card | |
| Misaligned Behavior in Realistic Work Environments | 2.28% | #2 of 3 | Model card | |
| BioSecBench-Surveillance | 39.9% | #20 of 27 | Official leaderboard | |
| TxBench-Oligo-Discovery | 30.4% | #22 of 28 | Official leaderboard | |
| OpenAI Deployment Simulation | 0.04% | #4 of 5 | Model card | |
| DistressBench | 65.9% | #35 of 42 | Official leaderboard | |
| Declining to Exploit a Honeypot During Difficult ExploitGym Problems | 0.07% | #5 of 5 | Model card | |
| Respecting Auto-Review | 0% | #5 of 5 | Model card | |
| Respecting Warnings | 29.6% | #5 of 5 | Model card | |
| Unintended Engagement with External Agent Messages | 65% | #5 of 5 | Model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 77.8% | #7 of 28 | Official leaderboard |
OpenAI's models over time
OpenAI's models in color, against every other model on the Epoch Capabilities Index, by release date.
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. 36 of them aren't in the launch post.
Health
Hallucination
Safety
Alignment
- Respecting Auto-Review 0
- Respecting Warnings 29.6
- Declining to Exploit a Honeypot During Difficult ExploitGym Problems 0.07
- Blocker Deception 8.47
- Broken Search Tool 1.83
- Misaligned Behavior in Realistic Work Environments 2.28
- Unintended Engagement with External Agent Messages 65
- OpenAI Deployment Simulation 0.045
Monitorability
Bio and chem risk
Cyber risk
- ExploitBench 97.1
- ExploitBench - Internal Port 24.5
- SEC-bench Pro 83.3
- Sandbox Bench 22.7