30 SEP 2026 — OpenAI released GPT-6.1 Sol at its DevDay developer conference on 29 September, one week after GPT-6 Sol. The standard price is unchanged at US$2 per million input tokens and US$10 per million output, and cached input now costs half as much. OpenAI says the new model nearly matches its flagship, GPT-6 Astra, on coding and computer use at a fifth of Astra's token prices.
The same event introduced "dots", always-on agents that run on Astra with their own cloud computers.
Same price, cheaper cache
GPT-6 Sol launched on 22 September at US$2 and US$10, half the previous rate. GPT-6.1 Sol keeps those prices. OpenAI's price list shows GPT-6 Astra at US$10 and US$50, five times as much.
The real change is cached input, the price charged when an application sends the same long instructions or documents again. It is now US$0.10 per million tokens, half GPT-6 Sol's cached rate and 95% below Sol's standard input price. For agents that replay the same context on every step, that line decides much of the bill.
What OpenAI says about performance
All the performance figures in OpenAI's announcement come from its own testing. On DeepSWE v1.1, a test of software-engineering tasks in real codebases, it says GPT-6.1 Sol matches Astra at about a fifth of the cost. On OSWorld 2.0, which tests agents operating computer applications, it comes within 2.1 percentage points of Astra at about a seventh of the cost per task.
The rival OpenAI chooses for comparison is Anthropic's Opus 5.5. On a scientific-workflow test, it says GPT-6.1 Sol averaged US$5.47 per task against US$23.21 for Opus 5.5 and US$23.80 for Astra, though Astra scored highest. OpenAI says its competitor figures come from publicly available reports rather than its own runs.
OpenAI also reports fewer factual errors than GPT-6 Sol on a set of deliberately difficult prompts, and lower failure rates on its agent safety tests, such as admitting when a search tool is broken.
Dots
Dots are the other headline. Each dot runs on GPT-6 Astra with its own cloud computer and browser, connects to more than 4,000 apps, and keeps working in the background. They are rolling out on the Pro, Business Premium and Enterprise plans in what OpenAI calls eligible markets, which it does not list.
OpenAI says dots start with rules about when to act alone and when to ask. Background research uses read-only access to connected apps, and "certain sensitive tasks, such as changing a password, always stay with you". A monitoring system can pause or stop a dot if it detects a safety concern, the company says.
The flagship that did not ship
The upgrade went to the cheaper model, not the flagship. A day before DevDay, OpenAI cancelled the October launch of GPT-6.1 Astra, Decrypt reports. Chief financial officer Sarah Friar told CNBC there are times the company will hold back. "When we have to pace the frontier, we'll do that. That's what we're showing right now," she said.
What it means for developers
For now, OpenAI's newest model is its mid-priced one, and its most capable model is the one powering the new agents. For developers, the question is whether the gap to Astra is worth five times the price for their work.
RECATOOLS is written with the assistance of Claude, made by Anthropic, whose Opus 5.5 is the rival model OpenAI compares against. Readers should weigh that interest.
View author profile → · Editorial policy
