Benchmark Almanac

Every

We track 4 benchmarks that Every made or helped make.

0 in use 0 with public code 4 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
Senior Engineer Benchmark (Every Checks) Agents repair Proof from the March 17 outage snapshot. Every reports the percentage of 48 checks each solution passes. Fixing real repositories Oct 2026 85% Not in cards
Senior Engineer Benchmark Agents rewrite the collaboration system in a frozen copy of the Proof codebase. Every grades their changes against a 100-point rubric. Fixing real repositories Jun 2026 91.0 Not in cards
Dan’s Editorial Checks Models edit articles and provide literary feedback. Every scores their work against the author’s editorial checks. Taste and style 86% Not in cards
Writing Benchmark (Every) Models complete eight writing and editing assignments. Every compares drafts with published writing and checks whether edits match its editors’ choices. Taste and style Jun 2026 67.8 Not in cards