OpenAI introduces GPT-6.1 Sol, a new mid-tier model that bridges the performance gap with GPT-6 Astra while significantly lowering costs, signalling a strategic shift in AI model segmentation and deployment.
OpenAI has released GPT-6.1 Sol, a mid-tier model in its GPT-6 family that sits between GPT-6 Luna and GPT-6 Astra and is intended to deliver near-flagship performance at a much lower price. According to OpenAI’s announcement and reporting from TechCrunch, the model went live on 29 September 2026 and is already positioned as the company’s newer Sol option in the API, effectively superseding GPT-6 Sol just seven days after that model’s debut.
The commercial logic is straightforward. OpenAI is charging the same headline rates as before for input and output tokens, but it has sharply reduced the cost of cached input, which matters most for agentic workloads that repeatedly revisit the same context. The company says the model can also handle long prompts, multimodal input and a broad tool stack through the Responses API, including web search, file search, image generation, code execution, hosted shell access, patch application, skills, computer use and MCP support. That makes it more relevant for software agents and document-heavy workflows than for simple chat tasks.
There are, however, important differences from GPT-6 Sol that will affect developers. OpenAI’s migration notes, as summarised by The Frontier and AI Nexus, show that GPT-6.1 Sol does not support the lowest reasoning setting that GPT-6 Sol offered, and it removes temperature, top_p and top_logprobs controls when reasoning is enabled. Tool calling now requires the Responses API, while Chat Completions remains available only for requests without tools. OpenAI has also said an Ultrafast mode is coming later, but it is not part of the initial launch.
On performance, OpenAI is making a clear claim that GPT-6.1 Sol narrows much of the gap with GPT-6 Astra while preserving a large price advantage. The company’s own charts, as described in the source material, show stronger results than GPT-6 Sol across several benchmarks, including coding, computer use and document analysis. OpenAI also says the model’s factual accuracy improved at lower reasoning levels, while independent testing by Artificial Analysis found GPT-6.1 Sol to be one point behind Astra on its Intelligence Index but materially cheaper per task. The same analysis said the model sat on the cost-versus-intelligence frontier at every effort level, suggesting it is particularly strong for users who care about spend as much as raw score.
Safety and rollout, meanwhile, remain central to the story. OpenAI’s system card addendum indicates that GPT-6.1 Sol is treated as critical in cybersecurity, high in biology and chemistry, and below high in AI self-improvement risk categories. The addendum also shows mixed results on misrepresentation and persistence measures, even as some misalignment indicators improved versus GPT-6 Sol. TechCrunch reported that GPT-6.1 Sol reached paid users in ChatGPT Work and Codex, including Plus, Pro, Business, Enterprise and Edu customers, while free users do not get access at launch. The broader context is that OpenAI is pushing a faster release cadence and a more segmented model line-up, with Luna, Sol and Astra now serving distinct use cases rather than interchangeable tiers.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
