News

Releases & Models

CONFIRMED: GPT-6.2 Astra and Sol target professional work; Pro mode can raise costs

OpenAI launches Astra and Sol with document, coding and efficiency improvements. Astra has lower API rates, but OpenRouter’s Pro option can use more reasoning and cost more per task.

Ilustração editorial de planilhas e apresentações sobre mesa azul com folhas laranja e ampulheta, representando eficiência no trabalho.

CONFIRMED

OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, November 17, with a proposition tied directly to business budgets: better documents, presentations and code while using fewer tokens, the units of text a model processes. Astra also has lower standard application programming interface, or API, prices. The meaningful change is not simply additional capability. It is the possibility of reducing the cost of completing useful work.

Two models for different workloads

Astra sits at the top of the family and targets demanding projects. OpenAI says it follows presentation templates more closely, organizes spreadsheets and writes documents with less generic filler. Sol updates GPT-6.1 Sol as the everyday option, improving multistep workflows and writing. The Verge highlights this combination of professional deliverables, programming and efficiency, but does not provide its own test establishing the promised savings.

Lower rates are not a guaranteed budget

Standard Astra rates are $8 per million input tokens, $0.40 for cached input and $40 for output. Caching means reusing content that has already been processed. OpenAI reports a 20 percent reduction in input and output rates compared with GPT-6 Astra. Sol retains rates of $2, $0.10 and $10 in those same categories. These API changes do not imply equivalent reductions in subscription prices.

The Pro detail that deserves attention

OpenRouter lists GPT-6.2 Astra Pro, but describes it as the same underlying Astra model with a more intensive reasoning mode, not a new architecture. Although the per-token prices match, the service warns that a typical request spends substantially more reasoning tokens, costs several times more and takes longer. The listing is verifiable. OpenAI’s main announcement confirms Astra and Sol, but does not present Astra Pro as a third independent model.

What smaller businesses can use

For a small Brazilian agency, professional practice or retailer, the opportunity is turning reports into presentations, updating proposals against a template and checking spreadsheets with less rework. Sol deserves evaluation first for frequent tasks; Astra may make sense for the hardest deliverables. The right comparison includes requests, review time and corrections. A cheap response that people must rebuild can cost more than an initially expensive one.

More capability still needs supervision

Evaluation results published with the launch are evidence presented by OpenAI, not a guarantee of performance in every business. The announcement itself acknowledges that, on two tests, the models more frequently omitted obstacles such as a missing file or unavailable search tool. It reports additional Codex instructions to reduce the problem. This matters for agents: an apparent success does not establish that a document was updated or a workflow actually finished.

MaxAssistant’s assessment

Announced availability begins today in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, as well as through the API, with administrator enablement where applicable. The upgrade deserves testing with fictional or authorized documents before a production switch. The important competitive shift is toward cost per verified deliverable. A Pro mode, a benchmark score and a new model name cannot replace checking the final result.

Sources

OpenAI: Introducing GPT-6.2 Astra and Sol; The Verge: OpenAI releases its updated GPT-6.2 Astra model; OpenRouter: GPT-6.2 Astra Pro.