Models
OpenAI promises nearly the same as its most expensive model for a fifth of the price
GPT-6.1 Sol arrives at a price five times lower than GPT-6 Astra and, according to OpenAI, performs almost as well. The figures behind that claim come from the company itself.
The essentials
2 confirmed facts · 2 according to the source · 3 open questions
- OpenAI unveiled GPT-6.1 Sol on September 29, 2026. It is an upgrade to GPT-6 Sol, the mid-range model in its GPT-6 family.12
- Used from other programs, it costs a fifth of what GPT-6 Astra does: $2 per million tokens (pieces of words) read and $10 per million written.12
- According to the sourceOpenAI says it performs almost as well as Astra at coding, operating a computer and professional work. The figures come from the company itself.12
- According to the sourceAccording to OpenAI, responses with at least one factual error fall from 11.4% to 7.7% in conversations chosen for being difficult, with reasoning set to low.2
Why it matters
The mid-range model is starting to do the expensive one's job. If OpenAI's figures hold up, anyone building products with AI can get a similar result while paying five times less. This suggests the competition is now being fought on price: Claude Sonnet 5.5, from Anthropic, has the same rate. But the comparison is made by the seller, and it leaves out that direct rival.
If you use ChatGPT Work or Codex, the model is already available there, according to MarkTechPost. The price cut applies to the rate paid by those who connect the model to their own programs.
The details
All the performance figures come from OpenAI's announcement, according to MarkTechPost2. On DeepSWE v1.1, a coding test, GPT-6.1 Sol matches GPT-6 Astra at about a fifth of the cost2. On science tasks Astra is still ahead, and OpenAI recommends it for the hardest research2. The model is already in OpenAI's API (the connection for programs), in ChatGPT Work and in Codex2.
GPT-6 Astra costs $10 and $50 per million tokens2. Claude Sonnet 5.5, from Anthropic, has the same rate as Sol, but OpenAI only compares itself with Claude Opus 5.5, which is more expensive, MarkTechPost notes2. Simon Willison, on his blog, repeated his homemade pelican test and saw no notable differences from the GPT-6 family3.
What we don't know
- Whether anyone independent gets the same results: all the performance figures come from OpenAI.
- How it compares with Claude Sonnet 5.5, which has the same rate: OpenAI only compares itself with Claude Opus 5.5.
- Whether it will come to other ChatGPT plans: the sources only mention ChatGPT Work.
Background
On October 2 we reported that Anthropic had made a similar move: Claude Sonnet 5.5, its mid-range model, comes very close, according to the company, to Opus 5.5, which costs twice as much.
Sources
- 1Introducing GPT-6.1 Sol
- 2OpenAI Releases GPT-6.1 Sol: Near-Astra Coding and Computer Use at One-Fifth of Astra’s Token Price
- 3GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price
Was this story useful? Yes Not really
This story was written in Spanish by an artificial intelligence system from the sources listed above, and translated automatically. Before publication, a program checks that every figure and every name appears in the sources, and that the translation keeps the same facts, figures and sources. It has not been reviewed by a person. Spanish original · How we work (in Spanish)