Itdaily - OpenAI launches GPT-6.2 Astra for professional work and Sol for everyday work

OpenAI launches GPT-6.2 Astra for professional work and Sol for everyday work

Text 'Astra GPT-6.2 Sol' on a dark background with star trails and a glowing orb.
Source: OpenAI

OpenAI launches two new AI models: GPT-6.2 Astra for professional work and GPT-6.2 Sol for everyday work.  

OpenAI is back with two new AI models: GPT-6.2 Astra and GPT-6.2 Sol. Last month, the AI company announced a whole series of new models, from GPT-6.1 Terra for efficient computer use to GPT-6 Cyber, which focuses on cybersecurity research. Now GPT-6.2 Astra and GPT-6.2 Sol follow. According to the company, they deliver better performance on demanding tasks with fewer output tokens, so you can do more within your usage limits.  

GPT-6.2 Astra for professional work 

OpenAI promises that slides, spreadsheets and documents from GPT-6.2 Astra feel less like AI text. They should also match the templates and professional conventions that a user supplies more closely. The model is also said to write cleaner code and build better frontends, especially for websites, games and 3D design. 

On Agents’ Last Exam, a benchmark for multi-step professional work, GPT-6.2 Astra scores 5.2 points higher than GPT-6 Astra, according to OpenAI, at less than a third of the cost. Compared with Anthropic’s Claude Fable 5.2, the difference is 5.5 points, at five percent of the cost. For Fable 5.2, however, OpenAI left out nine tasks in which a fallback model was used. 

OpenAI calls GPT-6.2 Astra the best model in the world for scientists, based on its own figures. On Terminal-Bench-Science 0.1, which measures how well a model carries out scientific work in a terminal, it scores nine percentage points higher than GPT-6 Astra. Here too, the cost is less than a third. The model also achieves the highest score on the Vals Science Index, in which a model interprets real biological data and designs experiments. 

Sol for everyday work 

According to OpenAI, GPT-6.2 Sol also beats the previous top model. On LifeVal, an evaluation of challenging everyday tasks, Sol scores higher than GPT-6 Astra, at roughly a seventh of the cost. Astra for Law, a configuration of GPT-6.2 Astra for lawyers, achieves an even higher score on the same test. 

Astra gets cheaper 

In the API, input and output tokens for GPT-6.2 Astra are twenty percent cheaper than for GPT-6 Astra. One million input tokens cost eight dollars, one million output tokens 40 dollars. Cached input costs 0.40 dollars per million tokens, a sixty percent discount. GPT-6.2 Sol keeps the API prices of GPT-6.1 Sol. For comparison: GPT-6.1 Terra costs one dollar per million input tokens. 

The models are available in ChatGPT Work, in Codex and via the API as gpt-6.2-astra and gpt-6.2-sol. All performance figures come from OpenAI itself. Independent tests still have to show whether the lower cost per task also holds up outside its own benchmarks.