A comprehensive catalog of every significant language model release across all major providers and sizes — from 135M edge models to 1.7T+ parameter frontiers. Tracks 239 models across 24 families and 12 providers.
In Q2 2024, GPT-4 was unchallenged. By Q1 2026, Qwen3.5-9B beats 120B models. DeepSeek R1 matched o1 at open-source cost. Llama 4 (10M context) is free. The frontier is now free.
o1-preview (Sep 2024) introduced chain-of-thought as first-class. By 2025, every provider had a reasoning tier: o3, Gemini Deep Think, Claude Extended, Grok-3, DeepSeek R1, QwQ. Inference-time compute = reasoning.
2024: models answered questions. 2025: they ran code, browsed web, controlled computers. Computer use (Oct 2024), OpenAI Codex agent (May 2025), Claude 4.5, GPT-5.4 native computer-use. The chatbot era ended.
Gemini dropped ~70% (Aug 2024). Mistral offered free API. DeepSeek R1 undercut o1 by 95%. GPT-4-level intelligence: ~$0.01/1K tokens by 2026. Claude Sonnet 4.6 beats Opus at $3/M.
2024: vision was experimental. 2025: every frontier model handles image, audio, video. Pixtral 12B (Sep 2024), GPT-4o native, Gemini 2.0 multimodal live API, Qwen3-Omni. Text-only = deprecated.