Data Changelog

We log every data or formula change so you can see what changed and when.

October 14, 2026

Phase 3: 23 new language models (Qwen 3.5, 3.6 and 3.8, Qwen3-Next, Qwen3-Coder 30B, Gemma 4, Mistral Small 4, Ministral 3, GLM-4.5-Air, MiniMax-M2, Kimi K2.6, DeepSeek V4 Flash) and 5 image and video models (FLUX.2, FLUX.1 Kontext, Z-Image Turbo, Qwen-Image-Edit, LTX-2). 29 new GPUs and Macs (including M5 Pro/Max, RTX Pro and Radeon AI PRO cards, P40, L4, H200, MI300X). New KV cache model for hybrid linear attention, shared-KV layers and compressed attention. Speed estimates now include per-layer overhead, which lowers mixture-of-experts figures to match measured llama.cpp results. 14 new GPU comparisons, 8 quantization comparisons and 13 new guides. Expanded About, Contact and Terms; longer guides with worked examples and troubleshooting; buyer advice, quality and image tables on comparison pages; model picks and setup tips on GPU pages.

October 12, 2026

Phase 2 launch: Can I Run It, fine-tuning, image and video, KV cache, GGUF size and tokens-per-second calculators; 10 GPU comparison pages; first 5 guides; 10 image and video model pages.

October 11, 2026

Launch: LLM VRAM calculator, 36 model pages, 39 GPU pages and 14 VRAM tier pages. Architecture values taken from each model's config.json on Hugging Face.