Phi-3-medium-4k-instruct-abliterated-v3: benchmark results

failspy's abliterated version of Microsoft's Phi-3-medium-4k-instruct, with the refusal direction orthogonalized out of the weights. Provider: Microsoft. Released 2024-05-22. Access: Open.

Unified ELO 1488 ± 1, rank #840 of 1553 rated models, from 7 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Open LLM Leaderboard - MuSR18.52Score91.2
Open LLM Leaderboard - BBH46.73Score86.6
Open LLM Leaderboard - MMLU-Pro37.78Score81.3
Open LLM Leaderboard - IFEval63.19Score75.3
Open LLM Leaderboard - GPQA8.95Score73.2
Open LLM Leaderboard - MATH Level 515.94Score63.7
Open Korean LLM Leaderboard32.5Average Score (%)25.2

Interactive version: theaggregate.ai/model?slug=phi-3-medium-4k-instruct-abliterated-v3 · How It Works · Data refreshed daily, snapshot 2026-09-08.