Leaderboard

Run the universal script with --publish — it scores your model, then submits the entry straight to the BenchLabs review queue. No PR, no manual file upload.

1. Get the script

curl -O https://benchlabs.ereneksi.com/script.py

2. Run it against your model

python script.py --model your-org/your-model --publish

3. Authenticate

No browser on this machine? You'll get a short code to approve from any browser (phone, laptop, wherever) — the same device-login flow as gh auth login. To skip that, pass --hf-token <token> (or set HF_TOKEN) to authenticate directly instead. Either way, your Hugging Face credential only proves who you are — it's never used to write anything.

4. Review & track

A Bench Labs moderator reviews the diff against the live board before it goes live. Track your submission, or submit via the browser instead of the CLI, at benchlabs.ereneksi.com/submit.

Notes

• Same model id: only the tiers you just ran are replaced — everything else is kept.
• 7-2026 runs report both modes: metrics.generative (exact_match · hybrid_score, alias-aware, per-item routing) and metrics.loglikelihood (acc · acc_norm · soft_score · soft_score_norm).
• model_revision + script_sha256 pin exactly what was evaluated.
• Text-to-image entries (FID / MS-COCO) live in a separate top-level t2i_models array in models.json — see add_t2i_schema.py for the entry shape.