Search public pages, research tools, and SAVRN solutions.
Organization
RLE-Bench
This profile connects the 3 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Robotics.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
instructions. 82 selected tasks across 5 environments. 72 recorded results and 10 pending tasks. Each evaluated task shows its latest result. Modified instructions are labeled and edits highlighted. Native outcome, execution, evidence validity and usage coverage are reported separately. Pending tasks have no synthetic scores.
Kinex Benchmark (Native Control Budget) Current defaults for new evaluations Environment Public action-step budget Control time LIBERO (all four suites) RoboTwin Official per-task 400–1,700 RoboDojo SIMPLE RoboCasa LIBERO action-step usage and limits on the site and in CSV exports exclude five environment initialization steps. Its runtime limit is 605
Kinex Benchmark Unified media and evidence storage for the Kinex Benchmark Space. The current index compares Codex + GPT-6 Astra / high and Kinex + GPT-6 Astra / high, with 84 selected seed-0 episodes per combination. Claude Code + Opus 5.5 / high and Kinex + Opus 5.5 / high are planned and have no results yet. Pending tasks remain pending; native verdicts, measured instructions, budgets and token counters are unchanged. Location data.json Current 164-task