Ant's Ling releases open-source Ling-3.1-flash: 560B-param MoE built for real-world agent workflows
量子位 · wechat · 2026-10-15
Ant Group's Ling team open-sourced Ling-3.1-flash, a 560B-param MoE (25B active per token) with hybrid linear attention (7 KDA + 1 Gated MLA per 8 layers), 1M context, and 512 routed experts. Quantization cuts weights from 534GB to 292GB, dropping deployment from 8 to 4 GPUs. It scores ELO 1673 on GDPVal-AA v2.1, intelligence index 41 on Artificial Analysis (210 tok/s, $0.19/M tokens), and hit 97.8% pass rate on a 17-hour in-house Lua compiler dev task. It powers Ant's internal desktop agent 'Xiaohu'; hands-on testing shows strong long-chain task performance but weakness with long-text attention and no native vision.
More from Models
- Why ChatGPT beats the better Claude with mainstream users: being first known — aitrendz_xyz · 2026-10-15
- 'You need high pain tolerance to review Gemini 4,' researcher quips amid error screens — gleech · 2026-10-15
- MiniCPM-V 4.7 runs a 35B MoE multimodal model on phones with just 3GB memory — nevrekaraishwa2 · 2026-10-15
- Gemini 4 Argon wrote 5,000+ lines of pure Python to build its own ray-tracing 3D engine — aitrendz_xyz · 2026-10-15
- JetBrains' thinking MoE Mellum 2.1 (12B-A2.5B) GGUF trends on Hugging Face — JetBrains · 2026-10-15
- Microsoft Accidentally Leaks OpenAI Secret: GPT-6.1 Sol Is GPT-6 Sol With Fewer Thinking Loops — evilsocket · 2026-10-15