Strengths
- OpenAI reports 57.9% on Terminal-Bench 4.0, a test of command-line work, ahead of Claude Fable 5.1 (55.8%) and Claude Opus 5 (52.6%).
- Good at operating a computer. OpenAI reports 72.6% on OSWorld 2.0, against 70.2% for Claude Opus 5.
- Reliable at finding things in long texts. OpenAI reports 96.3% on its MRCR v2 test with 512K to 1M tokens of context, while the older GPT-5.6 Sol scores 73.8%.
- Ranked 4th of 52 models on the independent Artificial Analysis Intelligence Index v5.0, the best OpenAI model on that list.
Weaknesses
- Expensive: $10 per million input tokens and $50 per million output tokens, five times the price of GPT-6.1 Sol.
- It does not win everywhere, even in OpenAI's own table. On Humanity's Last Exam with tools it scores 57.2%, Claude Fable 5.1 scores 65.0%.
- Slow to answer at about 73 tokens per second. Artificial Analysis ranks it 128th of 193 models for speed.
- Stricter cybersecurity checks can pause or stop legitimate work. It refuses advanced tasks such as proof-of-concept exploit code.
Best for
- the hardest coding and autonomous agent work
- agents that operate a computer or browser
- very long documents, up to about a million tokens
- professional and scientific research
Not ideal for
- high-volume simple tasks where cost matters
- chat where you need an instant reply
How to use it in Europe
In ChatGPT you get the most of it on Pro, where it powers the GPT-6 Pro mode. On Plus it is only available in ChatGPT Work and Codex, with limited usage. Companies get it on Business and Enterprise, where an admin has to switch it on. Developers use it through the OpenAI API, which is available in Slovakia and Czechia, or via Microsoft Azure and Amazon Bedrock. If you need data processed in the EU, regional processing in the API costs 10% extra. Eligible API customers can also use Zero Data Retention, where OpenAI does not keep the data.
What is coming
Rumor, unconfirmed According to The Information, OpenAI plans to release one or two new models before the end of the year. A major successor to GPT-6 Astra reportedly still needs safety work and may come later. When: one or two new models by the end of 2026 according to The Information, the flagship possibly later →
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index v5.0 | 57 (#4 of 52) | artificialanalysis.ai ↗, October 2026 |
| Terminal-Bench 4.0 | 57.9% | openai.com ↗, September 2026 |
| OSWorld 2.0 (offline) | 72.6% | openai.com ↗, September 2026 |
| GPQA Diamond | 96.0% | openai.com ↗, September 2026 |
| Humanity's Last Exam (with tools) | 57.2% | openai.com ↗, September 2026 |
Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.
More models from this vendor
- GPT-6.1 Sol: OpenAI's balanced model, close to GPT-6 Astra in quality at a fifth of the price. A solid pick for business apps.
- GPT-6.1 Terra: OpenAI's fast, cheaper model for working with images, operating a computer and everyday office work.
- GPT-6 Luna: OpenAI's cheapest model, made for high volumes of simple tasks, sorting text and low-cost chatbots.
- gpt-oss-120b: OpenAI's open model that a company can run on its own server when data must not leave its infrastructure.
- gpt-oss-20b: OpenAI's smaller open model that runs on a computer with 16 GB of memory, even without internet access.
Similar models from other vendors
- Claude Opus 5.5 (Anthropic): Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
- DeepSeek V4 Pro (DeepSeek): DeepSeek's largest open-weight model for demanding text tasks at a low price. It does not understand images.
- Gemini 4 Carbon (Google): Google's most capable model, announced on 28 October 2026. So far only selected cybersecurity experts have access.
- Gemini 4 Argon (Google): Google's most capable publicly available model, out in October 2026. Built for demanding analytical work where accuracy matters and made-up answers are costly.
- Muse Wonder (Meta): Meta's strongest model for long-running agents and coding, still released as a preview.
- Mistral Large 4 (Mistral AI): Mistral's largest model, still in preview. Built for demanding tasks, long documents and security analysis with data kept in the EU.