Simon Willison's PyCon US 2026 Recap: LLM Advances in Six Months

Simon Willison presented a lightning talk at PyCon US 2026 summarizing key developments in large language models (LLMs) from November 2025 to May 2026. He highlighted the November 2025 inflection point, where the "best" model changed hands five times among OpenAI, Google, and Anthropic. Starting with Claude Sonnet 4.5 (released September 29), the crown passed to GPT-5.1, Gemini 3, GPT-5.1 Codex Max, and finally Claude Opus 4.5. Coding agents improved dramatically, transitioning from often-failing to reliable daily-driver tools due to Reinforcement Learning from Verifiable Rewards by OpenAI and Anthropic. The open-source "Claw" personal AI assistant ecosystem emerged from the Warelay project (first commit November 2025), renamed OpenClaw by February 2026, driving Mac Mini shortages in Silicon Valley. In February, Gemini 3.1 Pro demonstrated advanced image generation, and Google's Jeff Dean tweeted an animated pelican riding a bicycle. Recent developments include Google's Gemma 4 open-weight models, Chinese AI lab GLM's GLM-5.1 (1.5TB parameters), and Qwen's Qwen3.6-35B-A3B, which outperformed Claude Opus 4.7 on a laptop. Willison concluded that laptop-available models now wildly exceed expectations.