Phi-4-reasoning-vision-15B: benchmark results

Provider: Microsoft. Released 2026-03-04. Access: Open.

Aggregate Rating 1488 ± 45, rank #2488 of 10516 rated models, from 31 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
UltraVR - Remote Sensing41.7Accuracy (%) on the remote sensing domain (comparison, logic80
UltraVR31.1Domain-balanced macro accuracy (%): mean of the CCTV, remote60
UltraVR - CCTV39.5Accuracy (%) on the CCTV surveillance domain (comparison, lo60
HarmVideoBench - Beyond-Clip Reasoning61.2Beyond-Clip Reasoning accuracy (%) on four-option multiple-c50
SciFigPlag-Bench - Reuse Localization3.3Paired localization accuracy (%; both the source-side and th46.7
LongShOTBench - Short Videos23.35Weighted Rubric Score (%)44.6
HarmVideoBench78.4Macro Avg. (self-reported)41.2
HarmVideoBench - Clip-Internal Meaning80.9Clip-Internal Meaning accuracy (%) on four-option multiple-c41.2
HarmVideoBench - Observable Evidence93Observable Evidence accuracy (%) on four-option multiple-cho41.2
PhysCount - Appearance NAE30.31Appearance normalized absolute error (%)35.9
LongShOTBench - Audio21.13Weighted Rubric Score (%)35.6
LongShOTBench - Long Videos22.6Weighted Rubric Score (%)35.6

Interactive version: theaggregate.ai/model?slug=phi-4-reasoning-vision-15b · How It Works · Data refreshed daily, snapshot 2026-11-21.