noviembre 18, 2026

OpenAI launches GPT-6.2 Astra and Sol: more professional capability and Astra’s API 20% cheaper

GPT-6.2 Astra y Sol

On 17 November 2026 OpenAI presented GPT-6.2 Astra and GPT-6.2 Sol. The first is its new frontier model for the most demanding projects; the second, the more efficient option for everyday work and multi-step agentic workflows. According to the announcement in the OpenAI developer community, both have been available since that day in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and in the API under the identifiers gpt-6.2-astra and gpt-6.2-sol.

The change companies already paying for the API will notice most is price. OpenAI says Astra’s input and output tokens are 20 % cheaper than GPT-6 Astra’s, with a 60 % discount on cached input: $8 per million input tokens, $0.40 cached and $40 output. Sol keeps GPT-6.1 Sol’s API prices. The company also says both models achieve better results on demanding tasks while using fewer output tokens, which further reduces the real cost per job.

What OpenAI promises — and how to read it

On its official blog, OpenAI says Astra improves the quality of presentations, spreadsheets and professional documents, writes cleaner code, and produces front ends — especially for the web, games and 3D design — that look and work better. On Agents’ Last Exam, a multi-step professional work benchmark, it claims Astra beats GPT-6 Astra by 5.2 points at less than a third of the cost, and leads Claude Fable 5.2 by 5.5 points at a fraction of the cost. These are OpenAI’s own figures, not those of an independent evaluator, and the company acknowledges that in that test Fable 5.2’s results exclude nine tasks where a fallback model was used. They should be taken for what they are: the manufacturer’s commercial case, useful for comparing orders of magnitude, not a certification.

Sol, for its part, is presented as the go-to model for everyday use. OpenAI says many of its own employees use it by default for both technical and non-technical tasks, and that on Cognition’s FrontierCode 2.0 it comes within 0.5 points of Astra at around a fifth of the cost — again according to its own charts.

What an SME can do with this

If you already use the OpenAI API, it makes sense to try Sol as the default model and keep Astra for critical documents, software architecture or analysis that justifies the price. If you were paying for GPT-6 Astra, check the model identifier: staying on the previous version is now more expensive and less capable. In any case, the savings only materialise if you measure tokens, pick the right model and avoid redoing work.

The other decision is about governance, not the catalogue. These models are also more capable in sensitive areas — OpenAI warns of stricter safeguards for higher-risk cybersecurity assistance. An SME should be clear about what data it sends them, who validates the output and what is logged.

At Digitacode we integrate these models into real processes, with cost and quality control, not as a standalone chat. If you want to update your integration or design a new one, write to us.

Sources