GPT-6 Luna is OpenAI's AI model, released 22/09/2026. GPT-6 Luna is OpenAI's high-volume low-cost model, priced at $0.10 input and $0.50 output per million tokens. OpenAI's launch post gives Luna scores only for DeepSWE and OSWorld, and the standalone GPT-6 Sol and Luna system card page could not be fetched, so the safety and health figures here come from Luna's comparison columns in the GPT-6.1 Sol and GPT-6.1 Terra system cards.
Visit GPT-6 LunaAI-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 Luna score on independent benchmarks?
GPT-6 Luna ranks #23 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 27.9, as measured on 22 November 2026.
Where does GPT-6 Luna rank against other models?
How good is GPT-6 Luna at agentic coding?
How does GPT-6 Luna do on individual tests?
| Evaluation (run by Artificial Analysis) | GPT-6 Luna | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 12.6% | 67.2% |
| FrontierCode 2.0 | 25.2% | 48.8% |
| Humanity's Last Exam | 38.5% | 74.3% |
| GDPval-AA (Elo) | 1443 | 1866 |
How has GPT-6 Luna's score changed over time?
We have tracked GPT-6 Luna 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 Luna system card report?
GPT-6 Luna is OpenAI's high-volume low-cost model, priced at $0.10 input and $0.50 output per million tokens. OpenAI's launch post gives Luna scores only for DeepSWE and OSWorld, and the standalone GPT-6 Sol and Luna system card page could not be fetched, so the safety and health figures here come from Luna's comparison columns in the GPT-6.1 Sol and GPT-6.1 Terra system cards.
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 Sol and Luna (22/09/2026), GPT-6.1 Sol system card addendum (includes GPT-6 Luna comparison columns) (29/09/2026), GPT-6.1 Terra System Card (26/10/2026), OpenAI API model page: gpt-6-luna. Last checked 22/11/2026.
Coding
| Benchmark | GPT-6 Luna | GPT-6 Sol | Claude Fable 5 | GPT-5.6 Luna |
|---|---|---|---|---|
| DeepSWE v1.1max effort | 66.6% | 68.8% | 69.9% | n/a |
Knowledge & science
| Benchmark | GPT-6 Luna | GPT-6 Sol | Claude Fable 5 | GPT-5.6 Luna |
|---|---|---|---|---|
| HealthBench Professional (length-adjusted)unadjusted 61.2%, mean 2,119 characters | 60.8% | 60.8% | n/a | 55.7% |
| HealthBench (length-adjusted)unadjusted 50.0% | 54.5% | 53.2% | n/a | 55.8% |
| HealthBench Hard (length-adjusted)unadjusted 25.4% | 31.4% | 30.1% | n/a | 32% |
| HealthBench Consensus (length-adjusted)unadjusted 95.6% | 95.9% | 96.2% | n/a | 95.1% |
| Multimodal Troubleshooting Virologypass@1, indicative High threshold 31% | 49.1% | 50.6% | n/a | n/a |
| ProtocolQA Open-Endedpass@1, indicative threshold 54% | 34.9% | 44.4% | n/a | n/a |
| Tacit Knowledge and Troubleshootingcons@32, indicative threshold 80% | 69.2% | 79.2% | n/a | n/a |
| TroubleshootingBenchpass@1, indicative threshold 36.4% | 38.6% | 45.3% | n/a | n/a |
Safety & honesty
| Benchmark | GPT-6 Luna | GPT-6 Sol | Claude Fable 5 | GPT-5.6 Luna |
|---|---|---|---|---|
| MentalHealthBench overallmaximum reasoning effort, standard error 0.9 | 51.7% | 54.2% | n/a | 44.4% |
| Cybersecurity Safety Evaluation, production chathigher is better; synthetic agentic environments 0.997 | 0.951 score (0 to 1) | 0.957 | n/a | 0.986 |
| Biology refusal evaluation, severe harm (not unsafe)higher is better; dual use 0.962, benign not over-refused 0.958 | 0.985 share | 0.998 | n/a | 0.946 |
| Adversarial coding deception rate (lower is better)OpenAI's test as charted in the launch post; Sol 1.3% (from 10.4%) | 2.8% | n/a | n/a | 9.5% |
What did safety testing find for GPT-6 Luna?
Safety level: OpenAI Preparedness Framework: per-model rating for Luna not stated in the documents read; evaluated against the Biological and Chemical High thresholds
- On biology, Luna scored 49.1% on Multimodal Troubleshooting Virology against an indicative High threshold of 31%, but below its thresholds on ProtocolQA Open-Ended (34.9% against 54%) and Tacit Knowledge (69.2% against 80%).
- Biology refusal evaluation: 98.5% safe on severe harm prompts and 96.2% on dual-use prompts, with 95.8% of benign prompts not over-refused.
- Cybersecurity safety evaluation scored 0.951 on production chat and 0.997 on synthetic agentic environments, slightly below GPT-5.6 Luna on production chat (0.986).
- MentalHealthBench overall rose to 51.7% from 44.4% for GPT-5.6 Luna, but stays below GPT-6 Sol (54.2%) and GPT-6 Astra (58.7%).
- OpenAI's launch charts show Luna's adversarial coding deception rate falling to 2.8% from 9.5% for GPT-5.6 Luna (reported by our article from the chart, hence site-article).
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our GPT-6 Luna 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.
