Scores come from 80 multiple-choice questions from SuperGPQA and MMLU-Pro. With that many questions, a score has a margin of error of several points, so models a few points apart are effectively tied on accuracy; the comparison pages run a significance test for any pair. To see the top two side by side, compare Granite 4.0 Micro and gpt-oss-20b.
Speed, latency and cost are measured through OpenRouter and depend on which host serves the request; your numbers on a direct API may differ. The full procedure is on the methodology page, and the data behind this ranking can be downloaded.
Which AI model gives the best accuracy for the money?
11 models sit on the frontier, from Granite 4.0 Micro ($0.000051 per question, 25.0%) to Claude Opus 5.5 ($0.015, 85.0%). Rule: The cost–accuracy frontier: each model listed scores higher than every model that costs less per question. Ordered from cheapest to most accurate.
What is the runner-up?
11 models sit on the frontier, from Granite 4.0 Micro ($0.000051 per question, 25.0%) to Claude Opus 5.5 ($0.015, 85.0%).
Which of these costs least per correct answer?
gpt-oss-20b: $0.00013 per correct answer (cost per question divided by its score).
Granite 4.0 MicroIBM
gpt-oss-20bOpenAI
Mercury 2.5Inception
GLM 5.3 FlashZ.ai
DeepSeek V4.1 Flash 1017DeepSeek
Gemini 3.8 FlashGoogle
Claude Opus 5.5Anthropic