Version 2.20.9

Submissions without files: choose your model and click Submit.

New

• You no longer download, edit or upload a file to make a submission. On the My Submissions tab you choose your model and click Submit; the platform builds the submission with your model's connection itself.
• Under Advanced settings you can change the temperature, max tokens and answer timeout. They come filled with the benchmark's own defaults, and only the settings that apply to that benchmark are shown.
• Benchmarks with image questions show the option to send them to the model, switched on.

Improvements

• The submission form shows the benchmark's answer mode (with reasoning or answer only) and whether the LLM judge is on. The organizer sets them; they are the same for every submission.
• Submissions are listed under the chosen model's name.
• The LLM judge address can be entered in both forms: the API base (https://api.example.com/v1) or the full call address (https://api.example.com/v1/chat/completions). A wrong address is refused with the accepted forms shown.

Fixes

• The model file no longer asks for the same connection details in two places and silently uses one of them; the connection has one source.