Ant Ling officially open-sources Ling-3.1-flash: supports up to 1M context, service deployment reduced to 4 cards

On October 15, Ant Ling officially open-sourced Ling-3.1-flash, enabling teams to evaluate the model's capabilities and inference efficiency within their own tasks and compute environments. Ling-3.1-flash has approximately 560B total parameters, activates about 25B per token, and supports up to 1M context. For local deployment, the team applied MXFP4 quantization to routed experts while keeping dense layers in block-FP8, reducing weight footprint from approximately 534GB to 292GB. In the corresponding Hopper GPU test configuration, service deployment was reduced from 8 cards to 4 cards, allowing teams with server-grade GPUs to run the model with fewer cards; actual requirements still vary with context, concurrency, and service configuration. According to the introduction, when Ling released Ling-3.1-flash on September 30, it already put the model into scenarios such as office work, medical communication, and software R&D. With this open-source release, Ling hopes more teams can deploy and integrate Ling-3.1-flash in their own environments, incorporate these scenario capabilities into their business processes, and serve real needs more flexibly. In Artificial Analysis's public evaluation, Ling-3.1-flash has an intelligence index of 41 and an output speed of about 210 tokens per second, providing dual capabilities of intelligence index and generation efficiency for high-frequency business scenarios. In Design Arena's Open Weights category evaluation, Ling-3.1-flash ranked third in all four directions: data visualization, React Native mobile apps, 3D design, and game development, and ranked fourth and fifth in SVG generation and website creation, respectively. Ling-3.1-flash can also assist in generating marketing pages, SaaS interfaces, React Native mobile app pages, and editable SVG assets based on text requirements, and can also create Web3D interactive prototypes and lightweight H5 games.