Today, we release LingSante-3.1-flash, an open-weight MoE model...

@AntLingAGI
Ant Ling@AntLingAGI
4 views Nov 22, 2026 ~2 min read
Advertisement
1
Today, we release LingSante-3.1-flash, an open-weight MoE model with 560B total parameters and 25B activations per token. It unifies diverse medical capabilities while retaining general capabilities, including complex medical reasoning, long-horizon workflow execution, and professional behavior alignment.
LingSante-3.1-flash moves beyond medical QA toward real-world healthcare problem. The model leads on HealthBench Professional (65.72), HLE (text-only, with tools: 72.22) and BrowseComp (with context: 94.10), outperforming GPT-6 Astra, Claude Opus 5.5 and Gemini 4 Argon on all three. LingSante-3.1-flash also performs well on professional long-horizon medical workflows, such as PhysicianBench (57) and EHR-Complex (74.40).
Together, these results establish LingSante-3.1-flash as the leading open-weight models for healthcare.
The weights are available. Explore LingSante and build with it.
More below ↓
Media image
2
Model Introduction

From sleep analysis to evidence retrieval and interactive consultation, LingSante-3.1-flash tackles multi-step healthcare tasks. It combines medical reasoning, tool use and context-aware guidance to turn questions into actionable outputs.
1:50
3
Complex Medical Reasoning

Sleep analysis needs traceable measurements. LingSante-3.1-flash computes heart rate and HRV by sleep stage in YASA. It preserves heartbeat timestamps for signal review and leaves metrics blank when data are insufficient.
0:42
4
Interactive Consultation

In a simulated case of headache and one-sided weakness, LingSante-3.1-flash flags danger signs. It updates recommendations as examination results arrive. The exchange becomes a structured medical record for review.
0:30
5
Evidence-to-Capability Matching
LingSante-3.1-flash uses ECM to match medical evidence to reasoning goals. MEDPAR refines tasks through multi-agent review. Samples are checked for medical correctness and task consistency before entering training.
Media image
6
Contribution-Aware Multi-Agent Search
LingSante-3.1-flash trains search agents with contribution-aware rewards. Against accuracy-only multi-agent RL, this improves accuracy while cutting total search steps by over 20%, balancing answer quality and cost.
Media image
7
Professional Behavior Alignment
LingSante-3.1-flash uses a behavior matrix for training rewards. Physician annotation agreement is 99.62%. It leads on HBP (65.72) and trails only GPT-6 Astra on the chart’s three-benchmark medical behavior average.
Media image
Media image
8
Long-Horizon Task Synthesis
LingSante-3.1-flash scales long-horizon tasks across real MCP, mock MCP and world-model environments. Tasks evolve across scenarios, tools and constraints. Cross-trajectory checks and task-level verifiers validate outcomes.
Media image
9
Try LingSante-3.1-flash via the API on Computrix. Start with a literature search or a multi-step research task.
Free API access to LingSante-3.1-flash on Computrix requires a US$20/month Basic Token Plan. If you only use the free version, eligible credits may be returned after the free period ends, subject to Computrix’s latest terms.
computrix.ai/sg/models/mode…
10
Run LingSante-3.1-flash in your own environment. Choose the original weights or the MXFP4 quantized version.
FP8
- Hugging Face: huggingface.co/inclusionAI/Li…
- ModelScope: modelscope.cn/models/inclusi…
MXFP4
- Hugging Face: huggingface.co/inclusionAI/Li…
- ModelScope: modelscope.cn/models/inclusi…
11
Building with LingSante? Join the Ant Ling Discord to share demos, ask questions, and compare notes with other developers.
discord.gg/jQtDsU5J6C
Actions
What You Can Do
  • Export as PDF or Markdown
  • Batch Export to Notion
  • Bookmark & Highlight
  • Screenshot Tweet
Create Free Account

Includes 7-day Premium trial

Advertisement