COMMUNITY · SUBMISSION

Submit a benchmark score

Ran a benchmark on your rig that we don’t have yet? Submit it here. Complete submissions enter review as hidden pending rows. They render publicly only after a first-party run or full verified review with log, reviewer, timestamp, and reproduction command.

BEFORE YOU SUBMIT

✓ Run the benchmark on your own hardware using our published runner (see methodology)
✓ Capture the full stdout + stderr and post to a public GitHub Gist (gist.github.com — required for the trust gate)
✓ Have the exact reproduction command ready (so a third party can re-run independently)
✓ Know which model slug from /models your run corresponds to (the catalog slug, not the HF repo name)

Benchmark*Which leaderboard does this score belong to?

Model slug*From /models — e.g. qwen-3-32b, turkish-llama-8b-instruct-v01

Quantization*Exact quant string — Q4_K_M, Q6_K, Q8_0, FP16, AWQ-INT4, etc.

Runtime*runtime-version — ollama-0.24, vllm-0.21, llama.cpp-b9283

Hardware*From /hardware — rtx-3090, rtx-4090, apple-m4-max, h100-80gb

Score (0-100)*The headline metric × 100. For pass@1 of 0.83, enter 83.0

Test-run Gist URL*https://gist.github.com/... — full stdout + stderr from the run. REQUIRED for public render.

Reproduction command*The exact CLI invocation you ran, including all flags.

Operator notes (optional)One paragraph if there’s anything notable — e.g. 'Q4 dropped 5pts vs Q8 on code; likely tokenizer rounding'.

Your handle*GitHub username or display name — credited on the row.

Contact (optional)Email or GitHub URL — only used if we have a verification question.

Read the trust contract first →