LongCat Flash Lite — benchmark results

Meituan LongCat's lite flash tier: 68.5B MoE with ~3B active params and a 256K context (early 2026). Provider: Meituan. Released 2026-01-29. Access: Open.

Unified ELO 1560 ± 27, rank #592 of 1776 rated models, from 38 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA TAU-2 Bench79.53Accuracy (%)69.6
ZeroEval MATH-50096.8MATH-500 Score69.4
Artificial Analysis Intelligence Index17.22Intelligence Index53.5
AA IFBench43.06Accuracy (%)46.4
AA-LCR25.7Score (self-reported)45.6
AA Terminal-Bench Hard10.61Accuracy (%)45.1
AA Humanity's Last Exam6.02Accuracy (%)44.1
AA GPQA Diamond63.64Accuracy (%)42.6
AA Omniscience - Software Engineering (SWE) - Dart18Accuracy (%)40.7
LLM Stats (CMMLU)82.48Score (%)40
ZeroEval GPQA Diamond66.78GPQA Diamond Score39.8
AA Omniscience - Software Engineering (SWE) - Kotlin14Accuracy (%)38.6

Interactive version: theaggregate.ai/model?slug=longcat-flash-lite · How the rankings work · Data refreshed daily, snapshot 2026-07-22.