Summary & Key Takeaways ​

• Simon Willison comments on the release of GPT 6.1 Sol, noting its 'near-Astra intelligence'.
• The model is highlighted for its claimed performance at a fifth of the price.
• He links to his live-blogging of the OpenAI DevDay 2026 keynote.
• Visual 'pelican' benchmarks for GPT 6.1 Sol are provided.
• The new model's pelicans are noted as not significantly different from the GPT-6 family.

Our Commentary ​

Simon Willison's 'pelican' benchmarks are always a fun, unique way to gauge LLM performance. While this is just a quick comment, it points to a potentially significant AI model release in GPT 6.1 Sol. The claim of 'near-Astra intelligence for a fifth of the price' is certainly attention-grabbing. We're seeing such rapid iteration in the AI space, it's hard to keep up, but these kinds of cost-performance improvements are what really drive adoption.