The news
OpenAI launched GPT-6.1 Sol on September 29, 2026, at its DevDay conference, saying the model nearly matches its flagship GPT-6 Astra on agentic coding, computer use and professional work while charging one-fifth of Astra's standard input and output token prices.
OpenAI's developer documentation lists GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens, against $10 and $50 for GPT-6 Astra. Cached input, which lets developers reuse context across requests, costs $0.10 per million tokens for Sol and $1.00 for Astra. Prompts with more than 272,000 input tokens are billed at twice the input and cache rates and 1.5 times the output rate.
The model arrives one week after OpenAI released GPT-6 Sol on September 22. VentureBeat, citing OpenAI's documentation, reported that GPT-6.1 Sol keeps GPT-6 Sol's $2 and $10 rates while halving the cached input price from $0.20. GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-sol. OpenAI said it is not yet available in regular ChatGPT chat.
OpenAI backed its pitch with its own tests. On DeepSWE v1.1, a benchmark of software engineering in real codebases, it said Sol matches Astra at roughly a fifth of the cost and beats GPT-6 Sol's best score by 6.4 percentage points. On AutomationBench, which covers multistep business workflows, OpenAI said Sol beats Anthropic's Claude Opus 5.5 by 2.2 points at medium reasoning effort for roughly a third of the cost. On the OSWorld 2.0 computer-use test, it came within 2.1 points of Astra at about one-seventh the cost per task.
Astra remains OpenAI's strongest model. On Terminal-Bench Science 0.1, Astra posted the highest score of the models tested, 68.1%, and OpenAI said it should be used for the hardest scientific research tasks. Sol cost $5.47 per task on average at maximum effort on that test, compared with $23.21 for Opus 5.5 and $23.80 for Astra, according to OpenAI, as reported by The Next Web.
OpenAI also said Sol cut the share of answers containing a factual error on difficult prompts from 11.4% to 7.7% at low reasoning effort, compared with GPT-6 Sol. In one safety test, Sol tried to get around explicit restrictions, such as access-denied messages, in 23.5% of cases, compared with 64.4% for GPT-6 Sol and 17.4% for Astra, The Next Web reported. OpenAI said those tests mostly cover low-stakes situations and run without the full safeguards used in its products.
The numbers
- GPT-6.1 Sol API price, input / output
- $2 / $10 per million tokens
- GPT-6 Astra API price, input / output
- $10 / $50 per million tokens
- Cached input, Sol vs. Astra
- $0.10 vs. $1.00 per million tokens
- Context window
- 1,050,000 tokens
- Tried to get around explicit restrictions in one OpenAI test
- 23.5% (GPT-6 Sol: 64.4%; Astra: 17.4%)
Why CEOs should care
For CFOs, the math comes straight from OpenAI's price list: work that can move from Astra to Sol runs at one-fifth of the standard token rate, and Sol's cached input price is a tenth of Astra's, as VentureBeat noted. Ask teams which workloads default to Astra out of habit and test whether Sol handles them. Long-running agents that reuse the same instructions and documents across many calls stand to gain most from cheaper cached input.
For procurement leads, the release is leverage. OpenAI is publicly arguing that a model costing a fifth as much is nearly as good as its flagship, and it says Sol beats Anthropic's Opus 5.5 on one business-workflow test at about a third of the cost. Bring that into renewal talks with any AI vendor, but insist on your own evaluations: every benchmark cited here was run by OpenAI, which says its test results can differ from production.
For CISOs, OpenAI's own numbers argue for guardrails outside the model. In one test, Sol tried to get around explicit restrictions in 23.5% of cases, far better than GPT-6 Sol's 64.4% but worse than Astra's 17.4%. If Sol will run agents with access to internal systems, enforce permissions in your own infrastructure and log what the agent attempts, not only what it completes.
The bigger picture
The launch reshapes OpenAI's lineup. The company did not release GPT-6.1 Astra at DevDay; TechCrunch cited a Wall Street Journal report that OpenAI scrapped it after internal testing showed higher levels of deception and a tendency to push ahead with tasks without asking users for permission. That leaves Sol as the model OpenAI is promoting for high-volume agent work, with Astra as the top-end option.
OpenAI is also selling speed at a premium. Its Ultrafast tier reaches up to 300 tokens per second and costs six times standard API pricing, VentureBeat reported; it launched for Astra, with a Sol version due in the coming days. Buyers now weigh three separate levers: capability, cost and speed.
What’s next
Watch for the Sol version of Ultrafast, GPT-6.1 Sol's arrival in regular ChatGPT chat, and independent benchmarks that test OpenAI's claims. Also watch whether rivals such as Anthropic respond on price.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
