A developer's personal coding benchmark ranks 19 models on 21 tasks
teortaxesTex · x · 2026-11-13
Developer aquilesfd built aequindex, a coding benchmark from his own multi-year projects: 21 tasks, 19 models, identical repo and tools, 348 points per task ceiling (Opus 5.5 high). The site plots scores against cost, time, tokens and behavior penalty with a 'most attractive quadrant'. teortaxesTex notes personal leaderboards can't be benchmaxxed and favor specified tasks over vague 'make me happy' prompts.
Related event: Developer Builds Personal Coding Benchmark Testing 19 Models on 21 Tasks(2 posts)→
More from coding & agent
- KubeCon demo scales agent swarm from 50 to 16 million actors in three days — rakyll · 2026-11-13
- MCP server fixes that appear out of nowhere? Your agent may have ratted you out — zeeg · 2026-11-13
- MIT & Meta paper: ReHarness revises agent memory notes with later outcomes, beats baselines — rohanpaul_ai · 2026-11-13
- Mnemos lands in Nous Plugin Marketplace: portable persistent memory for your agents — RileyRalmuto · 2026-11-13
- AI game platform data: ~70% play pop-in version, loading screens tank retention — TAbrodi · 2026-11-13
- Ridgeline chart of daily agent friction: prompts vs model pushback — dbreunig · 2026-11-13