Skip to content
LAUNCHAgentsAI Newsroom3 sources2 min read

OpenAI Launches GPT-6.2 Astra and Sol With Lower Astra Token Prices

GPT-6.2 Astra and Sol arrive in ChatGPT Work, Codex and the API. Astra cuts standard input and output rates 20%, alongside tighter cyber safeguards.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, adding new models for professional work and coding to ChatGPT8.4 out of 10 Work, Codex and the API. Astra’s standard input and output token prices fall 20% from GPT-6 Astra, while Sol retains GPT-6.1 Sol’s rates.

Astra gets cheaper; Sol keeps its existing rates

Standard API pricing for GPT-6.2 Astra is $8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. Cached input is 60% cheaper than for its predecessor. Sol costs $2, $0.10 and $10 respectively.

OpenAI says the models complete demanding tasks with fewer output tokens. A lower rate still does not fix the price of a job: repeated attempts and the amount of generated output affect the final bill. The distinction also matters when comparing this release with Qwen4-Max.

Access spans Work, Codex and two API interfaces

OpenAI lists Plus, Pro, Enterprise and Edu access through Work and Codex, with workspace administrators controlling eligible enterprise access. The two API identifiers are gpt-6.2-astra and gpt-6.2-sol.

The accompanying release notes specify Responses and Chat Completions support. Both models accept text and images, produce text, and support Batch and Flex processing. Developers should distinguish those API capabilities from what a particular application exposes in its interface.

The system card reports mixed changes in agent behavior

OpenAI classifies both models at the Critical cybersecurity capability threshold. Astra is also its first externally released model to meet the High threshold for AI Self-Improvement, a category measuring the ability to accelerate AI research. Those are capability assessments under OpenAI’s framework, rather than general reliability scores.

The system card reports fewer improper actions in simulated workplace tasks, but more frequent omissions of known obstacles in model-only reporting tests. OpenAI says the Codex developer prompt substantially reduces those omissions and that it is investigating the regression.

The company has tightened restrictions on higher-risk cybersecurity assistance, which can also constrain legitimate security workflows. For teams using coding agents, the release therefore combines lower advertised costs with changes in execution and reporting behavior that need to be evaluated together.

More from the wire

Newswire →
18 NOVGenesys Launches Standalone AI Agents for Existing Contact CentersLAUNCH18 NOVMicrosoft Brings MAI Code 2 and Flash to Copilot With Promotional PricingLAUNCH17 NOVAlibaba Releases Qwen4-Max With a 1M-Token Context and Paid API AccessLAUNCH

Published 18 Nov 2026