Ant's Ling-3.2-flash: 560B-param native multimodal model beats Opus 5 on release evals
alifcoder · x · 2026-11-06
Ant's Ling team introduced Ling-3.2-flash, a native multimodal model with 560B total parameters and 25B active per step, targeting complex visual understanding, long-horizon multimodal agents, 3D understanding and interaction, and medical intelligence.
Release evaluations claim wins over Opus 5 (max):
- OmniDocBench1.5: 92.21 vs 90.56
- ZEROBench(Sub): 43.71 vs 39.52
- Vision2Web: 85.86 vs 80.20
Pitched workflows include turning a single reference image into a detailed 3D scene, converting a research PDF into working code, and generating board-ready presentations from raw data. Note these are self-reported benchmarks.
More from Models
- Grok Bot on X Premium feels like a demo: user says usage pool can't last a week — nima_owji · 2026-11-06
- LlamaIndex teases ocria-1, claiming cheapest most accurate unified doc parsing model — _nerdai_ · 2026-11-06
- Can you run two Claude Ultra plans? Users weigh a ToS grey zone — 0xcydj · 2026-11-06
- Why no high-quant coding-pruned DeepSeek or GLM for 256GB machines? — evrenozkan · 2026-11-06
- ChatGPT's $20 price came from a 2023 Discord survey, and everyone copied it — aakashgupta · 2026-11-06
- ThinkyMachines' Inkling 2 Small trains models on its own: $426 of compute doubles Qwen score in 4.5 hours — Scobleizer · 2026-11-06