Benchmark Almanac

Akash

We track 2 benchmarks that Akash made or helped make.

0 in use 0 with public code 1 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
EvalCheck-LLM Models use supplied evidence to answer a question, abstain, or flag a conflict. The task tests whether pressure changes that decision. Faithful to sources Oct 2026 100% Not in cards
GhostInk: Honest OCR on Bill Photos Models read text from phone photos and mask unreadable characters. The grader penalizes guessed text, even when the guess happens to match. Documents, charts and text in images Oct 2026 0% Not in cards