TensorFold community leaderboard

Benchmarks

The models, the machines, and the people setting the pace.

Explore results

On the podium

Model capacity × measured speed from each contributor.

Parameter basis

Total model capacity × tokens per second. A sizing proxy, not quality or FLOPS.

Loading publisher model sizes

    Loading the community podium

    Across the community.

    Self-reported standard runs. Models and hardware differ.

    —published runs
    —contributors
    —models
    —hardware setups

    Across the community

    Model size and speed

    Publisher parameter estimates and measured token delivery.

    Standard runs

    Select a measured run to inspect its receipt.

    Apple SiliconNVIDIA CUDAVisible runs in this view

    Explore the benchmarks

    Every published run, newest first.

    Platform

    0 runs shown · Newest first

    All platforms

    Latest runs

    Select a model to inspect its receipt.

    ModelHardwareTokens / sFirst text msContributorReceipt
    Loading benchmarks

    Run a benchmark

    Chat, 256 output tokens, 5 repeats.
    Upload community-v1 or current community-v2 receipts. Choose the matching protocol and topology in the filters to compare runs; their rankings stay separate.

    Get the benchmark CLI
    tensorfold benchmark MODEL --output benchmark.json

    Run locally and keep the receipt. Uploads need no sign-in and accept community-v1 and current community-v2 receipts.

    Your community contribution

    Choose the public name shown beside your benchmarks. This browser remembers your contributor details.

    Display names are self-chosen and do not verify an identity.

    Review the public fields, then confirm to submit. Complete standard community-v1 and current community-v2 runs publish automatically after the model checks pass, within separate protocol and topology categories.

    Checked before publication

    Benchmark receipt