Leaderboards

Frontier general-purpose models, ranked on overall capability with a sanity check on price-per-million tokens.

Best models for autonomous coding agents and IDE assistants — pass rate on real engineering tasks first, latency and price next.

Models that crush math olympiad, scientific research, and multi-step logic problems. Quality dominates; cost is a tiebreaker.

Top picks for tool use, computer use, and long-horizon automation — judged on tool-call reliability and end-to-end task completion.

Vision, document, and screenshot understanding. Ranked on visual reasoning quality with multimodal context window as a bonus.

Long-form drafting, editing, and tone control. Quality of prose wins; cost matters once you start running drafts at scale.

Translation across mainstream language pairs, balanced 50/50 between fidelity and per-million-token cost.

High-volume support and chat — heavily price- and latency-weighted, but still required to get the answer right.

Always-on personal assistants and chat companions. Cheap, fast, and good enough for everyday Q&A.

Brainstorming, story writing, and ideation where stylistic range and originality matter more than raw price.