Best AI models for agentic tasks
Ranked by a 0–100 composite of agentic and tool-use benchmarks (Terminal-Bench, APEX, METR task horizon, GDPval (win/tie rate)), scored as percentile rank against every model released in the past year.
Agentic score is a 0–100 percentile composite across that area’s benchmarks. Open a model for the raw scores.
Frequently asked questions
What is the best AI model for agentic tasks?
Claude Mythos 5 (Anthropic) currently ranks first for agentic tasks on Model Beat, followed by Claude Fable 5 and Step 3.7 Flash. Ranked by a 0–100 composite of agentic and tool-use benchmarks (Terminal-Bench, APEX, METR task horizon, GDPval (win/tie rate)), scored as percentile rank against every model released in the past year.
Which is the most affordable strong agentic tasks model?
Among the top-ranked agentic tasks models, Step 3.7 Flash is the cheapest at $0.20 per million input tokens.
How are these agentic tasks rankings calculated?
Ranked by a 0–100 composite of agentic and tool-use benchmarks (Terminal-Bench, APEX, METR task horizon, GDPval (win/tie rate)), scored as percentile rank against every model released in the past year.
Benchmarks & model data from Epoch AI (CC BY); pricing & specs from OpenRouter. ECI = Epoch Capabilities Index.