OpenAI’s new GPT-6.1 Sol undercuts its own Astra flagship
The New Stack Frederic Lardinois ● Covered by 11 sources
OpenAI launched GPT-6.1 Sol, and it’s much closer to its pricey Astra model. It costs less, scores near the flagship, and may make Astra hard to justify.
Based on reporting by The New Stack, Frederic Lardinois — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI released GPT-6.1 Sol on Tuesday, just a week after GPT-6 Sol. The company is pitching it as an upgrade that gets very close to GPT-6 Astra on agentic coding, computer use, and professional work, while using only one-fifth of Astra’s standard token pricing.
The pricing itself hasn’t moved. GPT-6.1 Sol still costs $2 per million input tokens and $10 per million output tokens, with cached input priced at $0.10 per million tokens. It’s available now in the API and to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex. Not in Chat, though. OpenAI is also offering an Ultrafast version in Codex, with token generation up to 8x faster than the standard model.
The benchmark story is the real headline. Across most of the tests OpenAI shared ahead of launch, GPT-6.1 Sol lands very close to GPT-6 Astra despite the lower cost. On DeepSWE 1.1, a coding benchmark, it scores 6.4 percentage points higher than GPT-6 Sol and essentially matches Astra. OpenAI says the same pattern shows up in other tasks too, including complex PDF question answering, where GPT-6.1 Sol reaches about 32% on GDP.pdf versus roughly 29% for Anthropic’s Opus 5.5, again at about one-fifth of Astra’s cost per task.
Computer use is another strong spot. OpenAI says GPT-6.1 Sol beats its predecessor by seven percentage points at maximum reasoning and does so at half the cost, while staying in line with Astra. That is the kind of result that can make a flagship model look like a luxury, not a default.
There are some mixed results versus Sonnet 5.5, which launched on Monday at the same $2/$10 price. GPT-6.1 Sol scores around 75% on DeepSWE to Sonnet’s 71%, but trails on AutomationBench, where it posts about 36% against 44.7%. Even there, OpenAI points to a lower price per task: $0.30 versus $1.14. The company also says factual errors dropped, from 11.4% to 7.7% in a deliberately difficult test set, and that the model is better at following user intent and safety rules. In those tests, it didn’t try to work around a safety reviewer’s block at all.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that actually matters: cheaper, close enough, and harder to ignore. The industry loves a grand flagship launch, but the models that win are usually the ones that make the expensive sibling look a bit silly. Astra may still have a badge, but Sol is the one doing the sales damage.
Read more about this at: The New Stack