See what is available, which versions have independent results, and how an upgrade could change your bill. MiniMax M3.1 Flash Preview is awaiting independent results here; GPT-6.1 Sol and Claude Sonnet 5.5 have comparable benchmark results.
Tokens are the small pieces of text a model reads or produces. Include charged thinking tokens in output.
This estimate uses your token counts and standard text API rates. It excludes cache discounts, tool fees, retries and provider premiums. If your input and output exceed a model’s listed capacity, we leave its cost and Value rating blank. A large capacity does not guarantee that it will use every detail correctly.
THE MARKET MAP
Quality and cost
Sweet-spot zoneStrong price–quality tradeoffsTap a dot to inspect
Shaded zone: 75.0+ points and $21 or less. Numbers mark models with strong price–quality tradeoffs in or near this zone. On the dotted line, no other shown model is cheaper with an equal or higher score, or scores higher at the same price. Your chosen measure, usage and filters change this shortlist.
— means a comparable result is unavailable. Consensus combines results for only 10 matched configurations. Model IDs, reasoning settings, source links and pricing differences are available on every model page. Downloadable model files can still have restrictions on commercial use.
How production usage differs from benchmark quality
See which models Vercel AI Gateway customers use and where their token usage and spending go. Use those trends alongside task results and prices when building your shortlist.
There is no universal winner. Compare the task profile, tool support, reasoning configuration and total workload cost. Our Agent Fit score is a task-weighted shortlist; validate it using real tasks and your own tools.
How does AI Agent Store combine model benchmarks?+
Agent Fit gives more importance to the LiveBench skills needed for your selected task. Evidence Consensus combines relative rankings from LiveBench tests and Arena user votes only where the model version and setting match. Practical Value combines that ability score with how affordable the model is for your entered usage. Missing evidence is never filled with a guessed score.
Why do some models have no Consensus score?+
The snapshot has no verified matching Arena configuration for those models. Reasoning effort and version matter, so a result for a different setting is not silently reused. A missing score does not mean the model is weak.
Are these AI model rankings live?+
This is a reviewed snapshot, refreshed through a daily 05:30 Europe/Vilnius review. The last completed review is shown on the page; it is not a real-time feed. LiveBench uses its 25 June 2026 question set with model observations collected on the snapshot date. Arena and other sources have their own update dates, shown in the methodology.
What does an AI model context window mean?+
It is the advertised token capacity available to a request and its response. A large window lets you supply more information, but does not prove the model will retrieve or reason about every detail reliably.
Does a high benchmark score guarantee a reliable agent?+
No. Reliability also depends on prompts, tool design, permissions, retrieval, retry policies and evaluation. These pages compare models and reported evidence, not complete agent systems.
AI Agent Store