Changelog

Estimated monthly TCO per model + provider. Inputs: M tokens input/output/cached. Compares up to 12 models sorted ascending by cost. Cache support when the provider exposes it.

• → Pricing Calculator

4-step wizard to recommend the top 3 models per use case (coding, math, writing, vision, agent, RAG, summarization, translation). Composite = baseScore × coverage × priorityFactor.

• → Use Case Wizard

Side-by-side comparison of 2-4 models benchmark by benchmark. Shareable URL with ?models=a,b,c. Global verdict with winner by wins, abstentions when no score.

• → Battle Mode

Detects browser GPU/RAM/CPU (with honest Firefox/Safari limitations) and classifies each model into S/A/B/C/D/F tiers based on available VRAM and quantization. Multi-GPU 1-8.

• → Hardware Checker

Added total params, active params (MoE), license, cross-provider pricing and install commands (Ollama / LM Studio / vLLM) to the main open-weight models.

A benchmark is comparable only if 2+ models in the battle have a published score. If a model has no score, it counts as abstained (does not affect winRate).

• → Methodology

First public version of the atlas. 62 models, 32 benchmarks, 25 companies. Catalog, individual Models, Benchmarks, Companies, Methodology and Download single-file views.

• → Models