Technology

Add as a preferred source on Google

OpenAI Launches GPT-6.2 Astra and Sol With Lower Astra API Rates

OpenAI says Astra’s lower prices come alongside stricter cyber safeguards that may affect some vulnerability research and security-testing workflows.

Updated 2 min read
Pedestrians pass the glass-fronted offices of OpenAI in San Francisco on an overcast afternoon. / EconoTimes.ai
Pedestrians pass the glass-fronted offices of OpenAI in San Francisco on an overcast afternoon. / EconoTimes.ai

OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, pairing lower API charges for Astra with stricter cyber safeguards that the company says may constrain some vulnerability research and security-testing workflows.

Astra is aimed at demanding professional projects, while Sol is intended for routine tasks. At a volume of one million tokens, Astra’s rates are $8 for input, $0.40 for cached input and $40 for output. OpenAI said its input and output prices are 20% below GPT-6 Astra’s; cached input receives a 60% discount.

Sol’s rates remain unchanged from GPT-6.1 Sol: $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. OpenAI made both models available in ChatGPT Work and Codex to Plus, Pro, Enterprise and Edu users. Developers can access them through the API, and eligible Enterprise and Edu workspaces need an administrator to enable access.

OpenAI reported that Astra scored 5.2 points above GPT-6 Astra and 5.5 points above Claude Fable 5.2 on Agents’ Last Exam. The company estimated Astra’s cost per task at 5% of Fable’s and said Fable’s leaderboard score omitted nine tasks run with a fallback model. On Terminal-Bench-Science 0.1, OpenAI said Astra scored nine percentage points higher than GPT-6 Astra.

Independent coding results were less favorable. Artificial Analysis found Astra level with GPT-6 Astra on its Coding Agent Index, while Sol scored three points below GPT-6.1 Sol. The index measures coding performance separately from the evaluations OpenAI cited.

OpenAI’s system card says both models meet its “Critical” threshold for cybersecurity and its “High” threshold for biological and chemical capabilities. Astra also meets the “High” threshold for AI self-improvement; Sol falls below it. OpenAI said the tighter limits on high-risk cyber assistance are intended to reduce misuse, while acknowledging they may restrict some legitimate vulnerability research and security-testing workflows.

EconoTimes previously reported the identifier in Codex usage records before OpenAI specified a release channel; the launch confirms GPT-6.2 Astra’s route.

Related articles

View all