[AINews] GPT-6.2 Astra, the #2 Frontier Model is better and 20%-600% cheaper, but has mixed reviews; 6.2 Sol likely main workhorse
· Source: Latent Space
OpenAI announced the release of two new models, GPT‑6.2 Astra and GPT‑6.2 Sol, available on the ChatGPT platform (Plus, Pro, Enterprise, and Edu plans) and via the API under the names gpt‑6.2‑astra and gpt‑6.2‑sol. Compared to the previous version, Astra retains a one‑million‑token context window and can generate up to 128,000 tokens per response, while Sol keeps the same pricing structure as its predecessor (US$2 per million input tokens and US$10 per output token). Astra’s operating costs are reported to be 20 % to 600 % lower per task, according to tests by Artificial Analysis, and its Intelligence Index score is 62, just behind Fable 5.2 (65). In targeted evaluations, Astra showed gains in areas such as law (20.8 % accuracy versus 15 % for the base model), advanced mathematics, software engineering, and robotics, outperforming competitors like Fable and earlier GPT‑6 iterations. Social media reception has been mixed; some influencers criticize aspects of text generation and the launch presentation, though the company maintains that the model remains useful for automating tasks. This development is significant because it marks a step toward reducing the cost of high‑performance AI, potentially accelerating adoption by businesses and developers while intensifying competition among leading language‑model providers.
Read the original article on Latent Space
This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.