OpenAI introduced GPT-6.1 Sol on September 29 as an upgrade to GPT-6 Sol for coding, computer use and professional work. At launch, Plus, Pro, Business, Enterprise and Edu users could access it in ChatGPT Work and Codex, and developers could call gpt-6.1-sol through the API. OpenAI said the model was not yet available in the standard Chat surface.
Advertisement
Standard API rates are $2 per million input tokens and $10 per million output tokens, compared with Astra’s $10 and $50 rates. Cached input is a separate category at $0.10 per million tokens, half GPT-6 Sol’s cached-input rate. That distinction matters for agents that reuse context: the lower cache-read price applies to eligible cached tokens, not every input request. API pricing also does not define ChatGPT subscription allowances.
What changed beyond price
OpenAI says the model improves on GPT-6 Sol across software engineering, document work and computer interaction. On DeepSWE v1.1, a test of extended software-engineering tasks in real repositories, the company reports matching Astra while exceeding GPT-6 Sol’s best score by 6.4 percentage points at lower reasoning effort and cost.
The launch report also describes gains on professional PDF questions and multi-step business workflows. Those results explain the intended use more specifically than a general intelligence claim: reading complex documents, executing connected tasks and working through code changes. OpenAI’s evaluations use research or API configurations that can differ from production ChatGPT, and competing-model results come from public reports.
Astra remains stronger on some difficult tasks. On Terminal-Bench Science 0.1, covering scientific workflows such as data analysis and simulation, OpenAI reports Astra leading the tested models at 68.1%. It recommends Astra for the hardest scientific research work. Token-price ratios should likewise not be read as guaranteed reductions in the cost of completing every task.
Deployment and the next rollout
OpenAI’s system-card addendum classifies Sol as Critical for cybersecurity and High for biological and chemical capabilities under its Preparedness Framework. The company says it uses Astra’s safeguards. Those are OpenAI’s capability assessments and deployment controls, not independent assurance of every workflow.
At the announcement, OpenAI planned an Ultrafast option in Codex for the following days, claiming up to eight times faster token generation than standard speed. That plan is separate from the model’s initial availability. The release follows GPT-6 Sol and Luna, announced one week earlier, and chiefly changes the capability and cached-context economics of the Sol tier.