Ant open-sources LingSante-3.1-air-VL: 124B-A5.5B vision model claims medical SOTA
BerryBlush_Rogers691 · reddit · 2026-11-23
Ant's Ling team released and open-sourced LingSante-3.1-air-VL, a vision-language model with 124B total parameters, only 5.5B active, and 256K context, with weights on Hugging Face.
Per the official eval charts, it targets medical and document workloads: it tops DeepSeek-V4.1-Flash, GLM-5.3-Flash, Kimi-K3.1-Pro, Gemini 3.8 Flash, Claude Opus 5.5 and GPT-6.1 Sol on AFUMED-Derm (70.01), AFUMED-Report (90.70), CheXpert Plus (46.21), MIMIC-CXR (47.20) and OmniDocBench v1.5 (91.20), though it trails Kimi on BrowseComp-VL (62.80 vs 64.90) and GPT-6.1 Sol on AndroidWorld (86.84 vs 92.67).
More from Models
- Replit agents speed up NumPy 18.9x — but only Claude's team scales — amasad · 2026-11-23
- Perplexity is a flawed cross-lab capability bound: tokenizer granularity favors OpenAI — davidmanheim · 2026-11-23
- Gemini 4 Hydrogen at $1.50/M Input Tokens Scores 51, Beating Argon's 50 — dl_weekly · 2026-11-23
- Base log perplexity is an extremely strong predictor of Epoch and OpenAI ECI, researchers note — stochasticchasm · 2026-11-23
- Redis creator antirez: AI is already great at math and programming, blaming the tool is the real problem — antirez · 2026-11-23
- Gemini 4 xhigh costs ~2.8x more but passes the same 12 checks as high in coding test — mhdfaran · 2026-11-23