About

What TPS Leaderboard is

TPS Leaderboard is a community-built record of how fast local LLMs actually run — real tokens-per-second numbers from real hardware, submitted by the people running the models. Filter across six dimensions: OS, hardware, hosting tool, inference engine, GPU backend, and model. No sponsors, no cloud providers, no synthetic benchmarks. Every result is a claim the community can vote on, verify, and — when the numbers don't add up — flag. The goal is a leaderboard you can trust because you can see exactly how each number was earned.

Who's behind it

TPS Leaderboard is an independent, community-run project — built in the open for the local-LLM community, not by any hardware or model vendor.

Contact

Questions, corrections, or want to help moderate? Email hello@tokenspersecond.ai.

What's next

This is a passion project, built in spare hours, so no roadmap and no promises. Things we're tinkering with: a CLI harness that submits a run straight from your terminal, an in-app moderation queue, a submission API for CI rigs, and support for multi-GPU and speculative-decoding setups. If one of those matters to you — or something on the site looks broken — tell us on the issue tracker.