Ant Group's inclusionAI has released the weights of Ling-3.1-flash on Hugging Face under the MIT license. The model was announced on September 30 with a two-week free API trial, during which service was capped at a 256K-token context. The team said at the time that it planned to open the full 1M-token context and open-source the model once the free period ended and the service became paid. The free period closed on October 14, and the weights went up alongside it.
According to the official model card, Ling-3.1-flash is a native hybrid reasoning model built on Ling-3.0-flash, with 560 billion total parameters and 25 billion active per token. It uses a hybrid-linear mixture-of-experts architecture with 63 KDA linear-attention layers and 9 Gated MLA layers in a 7:1 ratio, 4 dense layers, 512 routed experts plus one shared expert, and 8 activated experts. The context window reaches 1 million tokens.
Ant says the model improves on general-purpose agents, search, office work and software engineering, and reports strong results on coding and agent benchmarks including Terminal-Bench 4.0, SWE-bench Pro, FrontierSWE, GDPVal-AA V2.1, Automationbench, MCP-Atlas and SkillsBench, as well as HealthBench Professional. The card says the model works well inside coding frameworks such as Claude Code, Kilo Code, Qwen Code, Hermes Agent and OpenClaw, and also cites gains in finance and materials science.
Deployment support arrived on day one. The card recommends SGLang, with a dedicated image supporting multi-token-prediction speculative decoding, and gives a reference server setup on eight H200 GPUs, or four GB300 GPUs. A vLLM recipe is also provided. The model is listed on ModelScope and OpenRouter as well.
The card also introduces LingCode, a harness built on the open-source coding agent Pi and tuned for long-horizon coding jobs that run for hours, involve hundreds of tool calls and require repeated context compaction. Ant says it is considering open-sourcing LingCode in the future.
Ling-3.1-flash follows Ling-3.0-flash, released in July, and the multimodal Ling-3.0-flash-VL, released in September.
SEE ALSO: Ant Group Releases 124B Ling-3.0-flash · Ant Group Opens Ling-3.0-flash-VL Multimodal Model