Back to Daily Feed 
GPT-6.2 Astra & Sol: Benchmarks, Pricing, and Analysis
Worth Reading
Originally published on Simon Willison's Weblog by Simon Willison
View Original Article
Share this article:

Summary & Key Takeaways
- The article analyzes OpenAI's new GPT-6.2 Astra and Sol models.
- GPT-6.2 Astra offers an 80% price reduction compared to its predecessor, with cached prices even lower.
- GPT-6.2 Sol maintains the same pricing as previous Sol versions, which were half the cost of GPT-5.6 Sol.
- Simon Willison updated his LLM tool to support the new models.
- Benchmarks were run using a complex Three.js animation prompt.
- Neither Astra nor Sol fully met the benchmark's creative requirements compared to other models.
- A detailed comparison grid and visual demonstrations of the benchmark results are provided.
Our Commentary
Simon Willison's immediate deep-dive into the new GPT-6.2 models is exactly what we need. The pricing changes are significant, especially for Astra, making longer agentic sessions much more viable. It's always good to see independent benchmarks, and his 'Yonkers mech' prompt is a fun, creative way to test. The fact that neither model fully nailed the creative aspect is a good reminder that even frontier models still have limitations. It's a balanced, insightful look.
View Original Article
Share this article: