AI Model Timeline

Extended thinking, improved agentic capabilities — powers Claude Code. Released Feb 5, 2026.

Frontier performance across coding, agents, and professional work. Released Feb 17, 2026.

Native multimodal reasoning + agentic tool use — Google's strongest model at launch

Efficient MoE with novel load-balancing and multi-token prediction — strong coding and math at low training cost

Thinking mode toggle — top open-source reasoning model

Tops most benchmarks + Chatbot Arena — 1M token context

o1-level reasoning via RL, fully open-weights MIT license — shocked markets

Best coding model at launch — outperforms GPT-4o on most benchmarks

Best open-weight model — competitive with GPT-4 Turbo on most tasks

Direct Preference Optimization — eliminates reward model and PPO from RLHF

Multimodal, passes bar exam, massive capability leap over GPT-3.5

Open-weights LLM that outperforms GPT-3 — enables community research

100M users in 2 months — AI enters mainstream consciousness

RLHF alignment via PPO on human preferences — precursor to ChatGPT

Few-shot in-context learning emerges at scale — no gradient update needed

Zero-shot multitask LM — "too dangerous to release" at launch

Bidirectional Transformer pre-training with MLM — GLUE SOTA across all tasks

Attention Is All You Need — self-attention replaces RNNs entirely