Compare two AI models on the same task

We start from the models we follow most closely and add the other models of the same generation from the same provider; the pairs below come from OpenAI, Anthropic, Google, DeepSeek, Qwen, xAI, Mistral, MoonshotAI and Z.ai. Batch variants, free listings, models no longer listed, models with too few checks so far and models whose name marks image, audio or music output are left out; dated snapshots of one model and variants at exactly the same price count once. Within one provider, each model is paired with the next tier of its generation by the cost of the L2 task, and the models we follow closely are paired with the next one from the same provider by that cost. Across providers, each of these models is paired with the model closest to it in L2 task cost from every other provider, when the higher of the two is at most twice the lower. For the providers most often searched for side by side (Anthropic and OpenAI, Google and OpenAI, OpenAI and xAI, DeepSeek and OpenAI, Google and xAI, Anthropic and DeepSeek and Anthropic and Google), each model we follow closely is also paired with the other provider’s closest model we follow closely, when the higher cost is at most three times the lower; for Anthropic, every pair of the Haiku, Opus and Sonnet tiers of one generation is included. The two names of a pair stand in alphabetical order; the order is not a ranking.

Models by provider and name. L2: deep analysis over a 50,000-token document in 3 steps; per month: 20 tasks an hour around the clock, 730 hours; visible output only, no cache discount, no hidden reasoning. Price your own task in the calculator. Common tasks: how many of the six tasks of the model picker fit on our data (published price, context window and the task’s rules); quality on your task is not measured by Measured Lab yet; — means the model is not in the picker.