Find the most useful open coding model that runs within 16 GB of system or GPU memory—and prove the result can be reproduced.
The reported system or accelerator memory must not exceed 16 GB for the submitted configuration.
Entries use the same public suite, dataset revision and output-scoring method.
A winning result needs independent evidence on equivalent or stricter hardware.
The starter’s seed metrics are fictional. Replace these entries before a real challenge.
| Rank | Entry | Evidence | Score | Memory | Speed |
|---|---|---|---|---|---|
| 1 | Orion Coder 7B with Ollama inside 16 GB open-lab/orion-coder-7b · RTX 3060 · 12 GB | Machine-captured | — | 7.2 GB | 31.4 tok/s |
Challenge definitions pin benchmark revisions, hardware rules, allowed model classes and tie-breakers.
Submission records reuse the standard Run manifest and verification workflow instead of creating a second data system.
Sponsor placement remains visibly labeled and cannot alter rankings, evidence badges or moderation decisions.
