GPT-6 Astra: benchmarks, price and rank
GPT-6 Astra from OpenAI is 5th of 123 rated AI models on Spezia's leaderboard, strongest on BrowseComp and FrontierSWE. It costs $10 in, $50 out per million tokens.
- 5th of 123 rated models
- Rating 1,927
- First on 3 of 17 benchmarks
- Closed
- Released Sep 3, 2026
- 1.1 million tokens of context
- Knows the world to April 2026
Where it is strongest
- 2nd of 56 on BrowseComp
- 2nd of 21 on FrontierSWE
- 1st of 6 on ARC-AGI-3
- 3rd of 17 on Remote Labor Index
- 2nd of 76 on FrontierMath
Where it falls behind
In the top half on every benchmark it has a result on.
Price: $10 in, $50 out per million tokens.
Its headline results
The benchmarks that count most in the rating, with its place among the models measured the same way, and where each number came from.
- Remote Labor Index20.8%
Real freelance projects that people were paid to do, from design and video to architecture, games and data work. The model gets the same brief and files the freelancer got and has to hand back finished work.
3rd of 17. Independent result from Epoch AI.
- APEX-Agents64.7%
Long tasks written by investment banking analysts, management consultants and corporate lawyers from their own work, done across files and apps in a realistic work setup.
14th of 56. Independent result from Epoch AI.
- Terminal-Bench 4.057.7%
Hard jobs done in a real computer terminal: building and fixing software, setting up systems, and science and engineering problems.
5th of 17. The maker's own number, from OpenAI, via LLM Stats.
- BrowseComp91.5%
Questions whose answers are hidden deep on the web and take many searches and pages to track down.
2nd of 56. The maker's own number, from OpenAI, via LLM Stats.
- FrontierSWE65.5%
Open-ended technical projects that take hours to tens of hours, from making systems faster to building large programs and applied machine learning research. Each run gets a 20-hour budget.
2nd of 21. Independent result from Epoch AI.
- Humanity's Last Exam54.8%
About 2,500 questions written by experts in more than a hundred fields to be as hard as they could make them, many needing a diagram or image.
Independent result from Epoch AI.
- ARC-AGI-399.9%
Interactive puzzle games the model has never seen, with no instructions. It has to explore, work out the goal and the rules, and win.
1st of 6. The maker's own number, from OpenAI, via LLM Stats.
- Vending-Bench 2$15,515
The model runs a simulated vending machine business for a year on its own: finding suppliers, ordering stock, setting prices and paying the daily fees.
2nd of 60. Independent result from Epoch AI.
- DeepSWE 1.174.1%
Software engineering tasks run in the same simple agent setup for every model.
9th of 27. The maker's own number, from OpenAI, via LLM Stats.
- SimpleBench83.6%
Multiple-choice puzzles about everyday situations, space, time and people, written to trip up models that pattern-match instead of think. Ordinary adults do well on them.
6th of 68. Independent result from Epoch AI.
12 more results in its full profile.
Every result with its source, what it costs per task, what jumped with its release, side by side with any other model, and answers to your questions about when to use it.
Models ranked near it
How the rating works, and every benchmark it uses: the model leaderboard.