Ling 3.1 Flash: benchmark results

InclusionAI's Ling-3.1-flash, a released hybrid-linear mixture-of-experts model with publisher-nominal 560B total and 25B active parameters and a 1M-token context window. The 1019 post-training revision and the base checkpoint remain distinct identities. Provider: InclusionAI. Released 2026-09-30. Access: Open.

Aggregate Rating 1788 ± 12, rank #112 of 10489 rated models, from 189 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
InclusionAI Ling 3.1 Flash Release - Browsecomp91.67Score (%)100
InclusionAI Ling 3.1 Flash Release - FrontierSWE75.16Score (%)100
InclusionAI Ling 3.1 Flash Release - Supermatbench (QA)51.23Score (%)100
Nejumi 4 - GLP - Expression99.67Score (%)99.3
OpenRouter Tau2-Bench Airline79.3Accuracy (%)99.3
Nejumi 4 - Toxicity - Fairness100Criteria met (%)97.8
Nejumi 4 - jaster (0-shot) - MAWPS99Exact match (%)97.5
Nejumi 4 - MT-Bench (Japanese) - Writing99.5Judge rating (1-10, x10)97.2
Nejumi 4 - MT-Bench (Japanese)98.5Score (%)96.7
Nejumi 4 - jaster (2-shot) - MAWPS99Exact match (%)95.6
Nejumi 4 - MT-Bench (Japanese) - Roleplay99.5Judge rating (1-10, x10)95
Design Arena (UI Components)1344Elo94.8

Interactive version: theaggregate.ai/model?slug=ling-3-1-flash · How It Works · Data refreshed daily, snapshot 2026-11-20.