Simon Willison Reviews Six Months of LLM Progress at PyCon 2026
Original: The last six months in LLMs in five minutes
Why This Matters
Documents rapid LLM progress and coding agent breakthrough in late 2025
Developer Simon Willison presented a five-minute lightning talk at PyCon US 2026 summarizing six months of LLM developments from November 2025. He highlighted the November inflection point when model leadership changed five times between OpenAI, Anthropic, and Google, and coding agents became reliable for daily use.
Willison's talk covered what he calls the November 2025 inflection point in LLM development. The 'best' model title changed hands five times between Claude Sonnet 4.5, GPT-5.1, Gemini 3, GPT-5.1 Codex Max, and Claude Opus 4.5. He used a 'pelican riding a bicycle' test to compare models. The key breakthrough was coding agents becoming reliable for daily use after OpenAI and Anthropic applied Reinforcement Learning from Verifiable Rewards. During the December-January holiday period, developers experimented extensively with these tools. Willison created micro-javascript, a JavaScript implementation in Python. A project called Warelay, first committed in November, evolved into OpenClaw by February, becoming a popular 'personal AI assistant' that sparked a new category called 'Claws,' leading to Mac Mini shortages in Silicon Valley.