PAL-Bench (PAL-TRACE) - Person Identity Resolution: leaderboard

Metric: Person identity resolution (0-1; harmonic mean of the share of a user's hidden social persons, 653 in all, that the profile names and the share of those bound to the correct face cluster; PAL-TRACE framework with the listed LLM as reconstruction backbone; 50 synthetic users, macro-averaged over users; Qwen3.6-35B-A3B semantic judge). Source: arxiv.org. Saturation forecast: Around June 2027. 9 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)0.56
2DeepSeek V4 Flash0.48
3GPT-5.40.48
4Claude Sonnet 4.50.45
5Qwen 3.6 35B A3B0.44
6Gemini 3.1 Flash Lite0.44
7GPT-5.4 Mini0.42
8Gemma 4 26B A4B (IT)0.4

Interactive version: theaggregate.ai/benchmark?slug=pal-bench-pal-trace-person-identity-resolution · How It Works · Data refreshed daily, snapshot 2026-09-26.