OpenAI unveils GPT-6.2 Astra and Sol with tighter cyber and bio safeguards

·

OpenAI on Tuesday launched two new models, GPT-6.2 Astra and GPT-6.2 Sol, saying they improve coding, computer-use tasks and efficiency while lowering costs. The release is also notable as a safety-policy update: OpenAI said the models cross higher thresholds for cyber and biological risk under its Preparedness Framework and are being deployed with tighter controls.

“We’re introducing GPT‑6.2 Astra, our new frontier model for your hardest projects, and GPT‑6.2 Sol, our efficient, versatile model for everyday work,” OpenAI said in its product announcement, “Introducing GPT‑6.2 Astra and Sol.” The company said the rollout starts immediately. “Starting today, GPT‑6.2 Astra and Sol are available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.” OpenAI also made both models available to developers through the OpenAI API. In simple terms, Astra is the company’s higher-end model for more demanding work, while Sol is the lower-cost general-purpose option.

The pricing underscores that split. OpenAI listed GPT-6.2 Astra at $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. GPT-6.2 Sol keeps the same pricing as GPT-6.1 Sol, OpenAI said: $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. For customers, that means Astra is positioned as the premium model for tougher jobs, while Sol is meant to handle everyday work at a lower price.

OpenAI also used the launch to make a performance case, though those results are company-reported and not independently verified in the announcement. On “Agents’ Last Exam,” OpenAI said GPT-6.2 Astra scored 5.2 points higher than GPT-6 Astra and 5.5 points ahead of Anthropic’s Claude Fable 5.2, while using less cost per task. On “Terminal-Bench-Science 0.1,” OpenAI said GPT-6.2 Astra scored 9 percentage points higher than GPT-6 Astra. The company said the broader aim of the update is to improve professional outputs, coding and computer use while increasing efficiency.

The safety disclosures are a central part of the release. OpenAI said both GPT-6.2 Astra and GPT-6.2 Sol meet the “Critical” cybersecurity threshold in its Preparedness Framework, the company’s internal system for classifying model risk. It also said both models meet a “High” threshold for biological and chemical capabilities, and that GPT-6.2 Astra meets a “High” threshold for AI self-improvement. OpenAI said it is deploying the models with its “strongest safeguards to date,” including tighter safeguards for higher-risk cybersecurity assistance. For more advanced cyber use cases, the company pointed users to Daybreak, its restricted-access program for vetted cybersecurity organizations. OpenAI said an accompanying system card includes more detail on results and methodology.

The launch builds on OpenAI’s GPT-6 lineup introduced earlier this year. The company released GPT-6 Astra in September 2026 as its flagship model focused on coding, science, computer use and cybersecurity, then later introduced GPT-6.1 Sol as a lower-cost model. Tuesday’s update packages the next step in that product family with a clearer safety message: OpenAI is not just offering a faster or cheaper model refresh, but pairing higher claimed capabilities with stricter handling rules for the kinds of cyber and bio assistance it considers more sensitive.

Tags: #openai, #ai, #cybersecurity, #machinelearning