OpenAI released GPT-6.2 Astra and Sol. Both models are more powerful, and Astra is cheaper

  • OpenAI released two new models: the more powerful GPT-6.2 Astra and the cheaper GPT-6.2 Sol
  • Astra is 20% cheaper in the API than GPT-6 Astra; Sol costs the same as its predecessor
  • In ChatGPT, they are currently only available in the Work environment and in Codex for Plus, Pro, Enterprise, and Edu subscribers

Sdílejte:
Jakub Kárník
Jakub Kárník
17. 11. 2026 23:30
Advertisement

A new model doesn’t always have to be more expensive. On Tuesday, November 17, 2026, OpenAI released GPT-6.2 Astra and GPT-6.2 Sol, and for the flagship Astra, it promises higher performance at a lower price. The company positions both models for professional work, i.e., science, law, finance, and programming. However, independent tests show that it’s not a leap forward everywhere.

Astra for demanding work, Sol for everyday use

According to OpenAI, GPT-6.2 Astra is a model for the most challenging projects. In the Agents’ Last Exam test, it scored 5.2 points more than the September GPT-6 Astra, at a lower price per task. The company designates it as the best model for scientists.

GPT-6.2 Sol builds on GPT-6.1 Sol and is aimed at general writing and programming. In the FrontierCode 2.0 programming test, according to OpenAI, it lags behind Astra by only half a point, yet it costs roughly one-fifth. Both models are designed to generate fewer output tokens for demanding tasks.

OpenAI did not release the GPT-6.1 Astra version due to security concerns, so the flagship series jumped directly to 6.2. However, even now, it’s not a completely smooth release: Astra is the first publicly available OpenAI model for which the company assigned a high risk in its security assessment regarding the AI’s ability to improve itself. The default cybersecurity safeguards are also stricter.

API pricing and where to find the models

In the API, GPT-6.2 Astra costs $8 per million input tokens and $40 per million output tokens. Sol remains at $2 and $10, respectively, just like GPT-6.1 Sol. A new feature for developers is the Ultrafast mode, which is designed to generate text up to 4.5 times faster but consumes more of the limit.

As of Tuesday, the models are available in ChatGPT Work and in Codex for Plus, Pro, Enterprise, and Edu subscribers. OpenAI is not adding them to the regular chat with this step. Furthermore, the rollout is gradual, so some users only see them on the web for now.

Independent tests are more modest

According to The Decoder, which summarized the results of independent measurements, Artificial Analysis gave Astra 62 points in its overall index. This is enough for second place behind Claude Fable 5.2 with 65 points. Sol has 57 points.

The catch is in programming. According to Artificial Analysis, Astra did not improve in its programming agent index compared to GPT-6 Astra, and Sol lost three points. Anyone choosing a model primarily for writing code should test it on their own tasks before switching.

Will you try Astra for work, or is the cheaper Sol sufficient for your everyday tasks?

Sources: openai.com, the-decoder.com, 9to5mac.com

About the author

Jakub Kárník

Jakub is known for his endless curiosity and passion for the latest technologies. His love for mobile phones started with an iPhone 3G, but nowadays… More about the author

Jakub Kárník
Sdílejte: