GPT-6.1 Terra is OpenAI's AI model, released 26/10/2026. GPT-6.1 Terra is OpenAI's mid-tier model, priced at half GPT-5.6 Terra. The launch post gives only a few exact scores (Terminal-Bench 4.0 and error rates on error-inducing prompts) and shows the rest in charts, so most figures here come from the Terra system card and the comparison columns in the GPT-6.2 Astra and Sol system card. OpenAI rates it Critical for Cybersecurity and High for Biological and Chemical, below High for AI Self-Improvement.
Visit GPT-6.1 TerraAI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does GPT-6.1 Terra score on independent benchmarks?
GPT-6.1 Terra ranks #9 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 51.3, as measured on 22 November 2026.
Where does GPT-6.1 Terra rank against other models?
How good is GPT-6.1 Terra at agentic coding?
How does GPT-6.1 Terra do on individual tests?
| Evaluation (run by Artificial Analysis) | GPT-6.1 Terra | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 55.6% | 67.2% |
| FrontierCode 2.0 | 42.3% | 48.8% |
| Humanity's Last Exam | 46.2% | 74.3% |
| GDPval-AA (Elo) | 1476 | 1866 |
How has GPT-6.1 Terra's score changed over time?
We have tracked GPT-6.1 Terra on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the GPT-6.1 Terra system card report?
GPT-6.1 Terra is OpenAI's mid-tier model, priced at half GPT-5.6 Terra. The launch post gives only a few exact scores (Terminal-Bench 4.0 and error rates on error-inducing prompts) and shows the rest in charts, so most figures here come from the Terra system card and the comparison columns in the GPT-6.2 Astra and Sol system card. OpenAI rates it Critical for Cybersecurity and High for Biological and Chemical, below High for AI Self-Improvement.
These are OpenAI's own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Sources: Introducing GPT-6.1 Terra (26/10/2026), GPT-6.1 Terra System Card (26/10/2026), GPT-6.2 Astra and GPT-6.2 Sol System Card (17/11/2026), OpenAI API model page: gpt-6.1-terra. Last checked 22/11/2026.
Coding
| Benchmark | GPT-6.1 Terra | GPT-5.6 Terra | GPT-6 Astra | GPT-6.1 Sol |
|---|---|---|---|---|
| Terminal-Bench 4.0max effort | 50.3% | 21.5% | n/a | n/a |
| SEC-Bench Pro (vulnerability reproduction)pass@1 | 83.3% | n/a | 85.4% | 78.8% |
| ExploitBench Internal Portmean arbitrary code execution, safeguards off | 24.5% | n/a | 31.5% | 21.5% |
| ExploitBench V8mean capability credit | 97.1% | n/a | 100% | 99.7% |
| Sandbox Bench5 of 22 targets solved | 22.7% | n/a | n/a | n/a |
Agents & tool use
| Benchmark | GPT-6.1 Terra | GPT-5.6 Terra | GPT-6 Astra | GPT-6.1 Sol |
|---|---|---|---|---|
| Internal Research Debugging (Initial)mean rubric score | 76.39% | n/a | 78.05% | n/a |
| KernelGen 1P (Initial)mean rubric score | 45.34% | n/a | 66.72% | n/a |
| NanoGPT (Secondary)mean baseline training time saved (indicative cutoff 80%) | 11.53% | n/a | 7.73% | n/a |
| PostTrainBench Lite (Secondary)mean performance gap closed (indicative cutoff 90%) | 60.54% | n/a | 76.97% | n/a |
| MLE-Bench Revised | 86.5 mean reference percentile | n/a | 93.8 | n/a |
Knowledge & science
| Benchmark | GPT-6.1 Terra | GPT-5.6 Terra | GPT-6 Astra | GPT-6.1 Sol |
|---|---|---|---|---|
| GPQA Diamond (no tools)accuracy | 93.56% | n/a | 96% | n/a |
| Multi-select Virology Troubleshootingpass@1 (indicative High threshold 31%); refusal-adjusted 52.54% | 46.52% | n/a | 55.78% | n/a |
| HealthBench Professional (length-adjusted)unadjusted 64.2%, mean 2,531 characters | 62.7% | 57.7% | 64.7% | 64.2% |
| HealthBench (length-adjusted)unadjusted 51.1% | 55.1% | 57% | 58.3% | 58.5% |
| HealthBench Hard (length-adjusted)unadjusted 26.0% | 31.4% | 32.7% | 36.6% | 36.2% |
| HealthBench Consensus (length-adjusted)unadjusted 94.8% | 95.1% | 95.1% | n/a | 96% |
| AAV Capsid Packaging (Critical threshold 0.60)higher is better | 0.5285 mean Spearman | n/a | 0.5282 | 0.5282 |
| SHP2 Function Prediction (Critical threshold 0.60)higher is better | 0.3696 mean R squared | n/a | 0.4 | 0.332 |
| Coronavirus-ACE2 Cell Entry (Critical threshold 0.75)higher is better | 0.4741 composite | n/a | 0.45 | 0.423 |
Safety & honesty
| Benchmark | GPT-6.1 Terra | GPT-5.6 Terra | GPT-6 Astra | GPT-6.1 Sol |
|---|---|---|---|---|
| Error-inducing prompts, error rate at max effort (lower is better)internal OpenAI test; 16.7% at low effort | 4.3% | n/a | n/a | n/a |
| MentalHealthBench overallmaximum reasoning effort, standard error 1.0 | 55.5% | 46.2% | 58.7% | 57.9% |
| Multiturn jailbreak, defender success at largest attacker budget | 69.5% | n/a | 67.2% | 63.6% |
| Cybersecurity Safety Evaluation, production chathigher is better | 0.98 score (0 to 1) | n/a | 0.97 | 0.987 |
| Biology refusal evaluation, severe harm (not unsafe)higher is better; dual use 0.960, benign not over-refused 0.982 | 0.99 share | n/a | 0.998 | 0.998 |
What did safety testing find for GPT-6.1 Terra?
Safety level: OpenAI Preparedness Framework: Cybersecurity Critical, Biological and Chemical High, AI Self-Improvement below High
- Uses the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra; reached 24.5% on ExploitBench Internal Port and 22.7% (5 of 22) on Sandbox Bench, with no successful honeypot exploits.
- In 29.6% of rollouts Terra saw a barrier but completed the task by an alternative route that violated the restriction, higher than GPT-6 Astra and GPT-6.1 Sol but lower than GPT-6 Sol.
- Made no attempts to bypass the auto-review monitor, against 7.3% of rollouts for GPT-5.6 Sol.
- Internal Codex deployment simulation produced 22 flags of severity 3 or higher (0.045%), against 24 for GPT-6.1 Sol, 38 for GPT-6 Sol and 61 (0.125%) for GPT-5.6 Sol.
- Verbalised evaluation awareness appeared in 9.88% of trajectories against 4.35% for GPT-6.1 Sol; MentalHealthBench overall rose to 55.5% from 46.2% for GPT-5.6 Terra.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our GPT-6.1 Terra deep dive.
OpenAI Model Timeline
1,050,000 tokens (max input 922,000, max output 128,000) context
1,050,000 tokens (max input 922,000, max output 128,000) context
1,050,000 tokens (max input 922,000, max output 128,000) context
1.05M tokens context
1,050,000 tokens (max input 922,000, max output 128,000) context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
