PPolyCodeBenchPublic release explorerRepository ↗

Audit reports Verify attestation Submit model Curator access

Loading release data

Connecting to the public release API…

PolyCodeBench public release explorerEvery result is tied to a versioned release.

Reviewed requests

Request a model evaluation

Submit model and source details for review. A request stays pending until an authorized administrator verifies the source, endpoint, model configuration and a fixed run budget.

Current published release context

Release v5

Exploratory release · published

Scope exploratoryVersion 5Published 2026-10-09

Release limitations

Live exploratory release: real model runs scored by the local pipeline; unranked, not calibrated and not a ranked benchmark result.
Built from 4 persisted scorecard(s) over 12 attempt(s) in 1 run(s); unscored attempts reduce coverage and are never imputed.
8 attempt(s) ended in a model failure (invalid or unparseable model output) and are scored as zero, not treated as missing coverage.
The scoring policy and language profiles behind these scorecards are not calibrated, so scorecards stay in needs_review and no total score is computed; read the per-task outcomes and coverage, not a single number.
Uncertainty intervals, judge calibration, scorer replay, native labels and rights review were not performed; they are recorded as not applicable to this exploratory release.
Task identities and content are not disclosed; entries cannot be compared as a rank.

Frozen methodology · live-exploratory-v1

Request review

Only model metadata, public endpoint/source URLs and the rights attestation are sent.

A request creates no model call, VM, benchmark run or charge. An administrator must verify source rights, approve a configured endpoint, and bind one model configuration, task set, run plan and fixed budget before a queued run can be created.

Track your request

Only the verified account that submitted the request can read its status.