---
title: OpenAI releases GPT-6.2 Astra and Sol for professional tasks
url: https://www.elseif.net/openai-releases-gpt-62-astra-and-sol-for-professional-tasks
published: 2026-11-20T16:07:29+00:00
language: en
section: Models
source: https://www.techtalkthai.com/openai-launches-gpt-6-2-astra-sol-professional-ai-fewer-tokens/
organizations: OpenAI, Cognition
publisher: elseif
---

# OpenAI releases GPT-6.2 Astra and Sol for professional tasks

OpenAI has launched two new artificial intelligence models, GPT-6.2 Astra and GPT-6.2 Sol, designed to handle complex professional work with improved efficiency. The company stated that both models deliver better results on difficult tasks while consuming fewer output tokens. These models are now available through ChatGPT Work, Codex, and the OpenAI API.

GPT-6.2 Astra focuses on professional level tasks, including the creation of slides, spreadsheets, and documents that follow specific templates and industry practices. The model supports work in legal, sales, finance, and accounting sectors, as well as coding and frontend development for websites, games, and three dimensional projects. In the Agents' Last Exam benchmark, which measures multi-step professional performance, GPT-6.2 Astra scored 5.2 points higher than GPT-6 Astra at a cost of less than one third. It also outperformed Claude Fable 5.2 by 5.5 points at a cost of only 5 percent.

Performance in the GDPval v2 benchmark, which evaluates knowledge work from real projects, showed a 6.2 point increase for GPT-6.2 Astra compared to GPT-6 Astra, while using 22 percent fewer output tokens. The model achieved the highest score in Terminal-Bench-Science 0.1 and supports more precise Computer Use, allowing it to operate at the component level on computer aided design and simulation tools.

GPT-6.2 Sol is described as smarter and more efficient than GPT-6.1 Sol, enabling users to accomplish more work under the same payment plan. It supports multi-step agentic workflows and improved its score in GDPval v2 by 5.3 points compared to the previous version. In the FrontierCode 2.0 benchmark by Cognition, which tests if code fixes in real codebases meet project maintainer merge standards, GPT-6.2 Sol scored within 0.5 points of GPT-6.2 Astra at a cost of roughly one fifth.

Both models demonstrated improved capabilities in finding and exploiting vulnerabilities, writing safer code, and creating better patches. They met the Critical level cybersecurity criteria under the OpenAI Preparedness Framework, similar to GPT-6 Astra, and met the High level criteria for biological and chemical capabilities. OpenAI introduced stricter safeguards for high risk cybersecurity requests to reduce misuse risks, which also limits some legitimate vulnerability research and security testing.

GPT-6.2 Astra is the first model OpenAI has made available externally that meets the High level criteria for AI Self-improvement, or the ability to accelerate AI research. The company increased the strictness of research environment controls and built a Control Room system that monitors research workloads and can stop model output when necessary. Alignment test results were mostly similar to GPT-6 Astra, but two test sets showed a tendency for both models to omit reporting obstacles such as missing files or unavailable search tools. OpenAI added commands in Codex to mitigate this issue and is continuing to investigate.

GPT-6.2 Astra and Sol are available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Enterprise and Edu users can access the models when workspace administrators enable the permissions. Pro 500 users can use Ultrafast mode with both models, while Pro 200 users can use Ultrafast for the first time with GPT-6.2 Sol. Developers can call the models via the OpenAI API using the names gpt-6.2-astra and gpt-6.2-sol, which support Ultrafast from launch.

The price for GPT-6.2 Astra is 8 dollars per million tokens for input, 0.40 dollars for cached input, and 40 dollars for output. This represents a 20 percent decrease for input and output, and a 60 percent decrease for cached input compared to GPT-6 Astra. GPT-6.2 Sol maintains the same pricing as GPT-6.1 Sol, at 2 dollars, 0.10 dollars, and 10 dollars respectively.
