HOT RADAR
llmHEAT 16.6°

AI CURATED HOT STORY · 11/23/2026

Ant Group Open-Sources 560B Medical MoE Model LingSante-3.1-flash

1 reports archived1 independent sourcesupdated 11/23/2026, 03:22:49
Synthesis & Latest Updates
1 Sources Cross-Validated

Ant Group has open-sourced LingSante-3.1-flash, a 560B-parameter medical MoE model with 25B activations per token, releasing weights on Hugging Face under an MIT license. It outperforms GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon on HealthBench Professional, HLE, BrowseComp and EHR-Complex. The recipe mixes medical and general data from pretraining onward, trains medical reasoning, tool use, search and clinician-style behavior as separate skills with their own RL rewards, then merges the specialists via multi-teacher on-policy distillation.

LATEST/Ant Group has open-sourced LingSante-3.1-flash, a 560B-parameter medical MoE model with 25B activations per token, releasing weights on Hugging Face under an MIT license. It outperforms GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon on HealthBench Professional, HLE, BrowseComp and EHR-Complex. The recipe mixes medical and general data from pretraining onward, trains medical reasoning, tool use, search and clinician-style behavior as separate skills with their own RL rewards, then merges the specialists via multi-teacher on-policy distillation.

Hourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first
  1. Rohan Paul (@rohanpaul_ai)T2·88 pts
    • 560B total / 25B active MoE with MIT-licensed open weights available for download
    • Beats GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon on HealthBench Professional, HLE, BrowseComp and EHR-Complex
    • Recipe mixes medical and general data from pretraining, applies per-skill RL, then merges specialists via multi-teacher on-policy distillation