About TOKRACE
An independent, open-source model speed testing tool for developers, AI reviewers, and teams choosing models.
Updated: September 9, 2026
What we measure
TOKRACE sends the same prompt to multiple models in parallel and records time to first token, reasoning and output throughput, peak speed, and token usage. It helps turn impressions about speed into results that can be checked and repeated.
Speed is one part of model selection. These measurements do not establish that a model is more accurate or better at every task. Use your own tasks and review the outputs alongside latency and cost.
Data and methodology
Leaderboards aggregate measurements voluntarily contributed by users. Network conditions, endpoints, load, and sample sizes affect results. Pricing pages link to provider sources; the provider's current price prevails.
Read the measurement methodology · Speed leaderboard · API pricing
Independent and open source
The site is maintained under the TOKRACE name by the Model-Arena open-source project maintainers. TOKRACE is not an official ranking or an endorsement from any model provider. Provider names and logos identify the services being compared.
Advertising, sponsorships, or affiliate relationships, when present, are labeled. We do not sell ranking positions or change measurements in exchange for payment.