Version 2.21.0

2026-10-24

A major update: submissions without files, correct scoring of Turkish, one answer rule, protected model keys and a fixed record of every run.

This version brings together every update from 2.20.7 to 2.20.14. The details are in the release notes below.

New

  • Submissions without files: on the My Submissions tab, choose your model and click Submit. You never download, edit or upload a file; submissions are listed with your model's name.
  • Advanced settings: temperature, maximum tokens and the response timeout can be changed in the submission form; benchmarks with image questions send the images to the model.
  • Run record: every run's id and its fixed record (model, data, prompts, scorer, settings) are shown on the Run record tab of the submission details and can be downloaded as JSON. If the model changes during a run, the run is marked Not comparable.
  • One answer rule: the model's reply is read by the same published rule in every benchmark (the Answer Format section); JSON replies are read too. The model's own reply is kept for every question.
  • Twenty Questions: every player plays against the same fixed answerer; the secret word never reaches the model. The score formula and the time limits are on the rules page.
  • Image benchmarks: questions are grouped into categories; each category's weight and question count are published.

Improvements

  • The reason a submission failed is shown briefly and clearly.
  • A model that cannot run in this environment is reported, with the reason, before the submission starts.
  • A model key never appears in a file or on a screen; submissions need no key. Every key can call only its own model.
  • No English text is left in the Turkish interface; terms are used consistently. Empty screens show the next step.
  • The model upload form explains how to prepare the archive; a wrong archive is rejected with the reason.

Fixes

  • Turkish answers are scored correctly: upper and lower case (İ/i, I/ı), Turkish quotation marks and apostrophes no longer turn a right answer wrong.
  • Undecided replies (A or B), and questions where the model could not be run, count as unanswered; no points by chance.
  • Security updates: the 21 October 2026 Node.js security release is applied to every component; for models, the platform calls only its own model service.
SSB Logo SSYZ Logo SAYZEK Logo

SSYZ Benchmark Platform

SSYZ Benchmark Platform is an enterprise platform that enables secure and comparable evaluation of large language models on the SSYZ network infrastructure operated under SSB coordination.

Presidency of Defense Industries - Artificial Intelligence Center (SSYZ)

www.ssb.gov.tr ssyz.org.tr www.sayzek.org.tr Release Notes
Edit Created By: Key: Is Public: Yes No Files Solutions Search by name... Show Public Tasks Upload Task Create Task Delete Selected Tasks Name Description Creator In Use Public Actions Select Users Select top 5 biggest users Download as CSV Last snapshot date: No snapshot has been taken yet User Joined at Datasets Submissions Total usage Benchmarks usage Users usage Administration usage Orphaned files usage Select top 5 biggest benchmarks Benchmark Organizer Creation date Submission Management Search Users... Search Benchmarks... Re-run Selected Stop Selected Delete Selected File Name File Size Submitted at Processing Start Processing End Compute Worker Queue Docker Image User Management Portal Search Users... Name Username Email User Type Delete User Delete User ? Cancel Change User User ID Switch User Search... Delete Selected File Name Benchmark Size Uploaded Delete? No Datasets Yet! Details Key Created By Created Description: Used by: Download File Close Filter Leaderboard by Columns Start typing to filter columns under 'Meta-data' or Tasks. Task: # Participant Date ID This CSV This JSON You are not a participant of this benchmark. Please register in My Submissions tab to view the leaderboard. Your request to participate in this benchmark is waiting for an approval from the benchmark organizer. Your request to participate in this benchmark is denied. Please contact the benchmark organizer for more details. No submissions have been added to this leaderboard yet! Preparing submission... this may take a few moments... Scoring Ingestion ---- Approved Pending Denied Unknown Show deleted accounts Email all participants Download all participants Username Is Bot? account deleted Revoke Approve Send Message Send Email Subject A message from the admins of {competition_title} Content Number of submissions used for the day The submission limit resets at midnight server time. Number of total submissions used All ID # File name Owner Phase Date Status Score Detailed Results Actions Run Submission Rerun Submission Delete Submission Cancel Submission Add to Leaderboard On the Leaderboard Make Public Make Private Task Admin ERROR: Submission is a parent, but has no children. There was an error during creation.