On October 15, Ant Group's Bailian large model announced the open-sourcing of Ling-3.1-flash. The model focuses on "real usability," achieving capability upgrades for real-world tasks so that vague business needs can more quickly become usable results. The open-source model's weights are available for download on Hugging Face and ModelScope.
Ling-3.1-flash continues the hybrid linear architecture of Ling-3.0-flash, with about 560B total parameters and roughly 25B parameters activated per Token; it supports up to 1M context. The model maintains high-quality performance in complex workflows such as office work, healthcare, software engineering, and terminal operations. Faced with scattered materials, it can help organize reports and presentation materials; faced with fragmented symptoms, it helps sort out medical history and consultation questions; faced with legacy code, it can understand the original implementation, write new features, and continue making corrections based on test results.
In the office task benchmark GDPVal-AA V2.1, Ling-3.1-flash scored 1673 (ELO). In the HealthBench Professional healthcare benchmark, the model scored 65.35. In the software engineering benchmark FrontierSWE and the terminal operations benchmark Terminal-Bench4.0, the model likewise demonstrated the ability to handle complex engineering tasks.
Through this open-sourcing, Bailian hopes that more teams can deploy and integrate Ling-3.1-flash in their own environments, incorporating these scenario capabilities into their own business processes.
When facing real tasks daily, enterprise knowledge assistants and coding assistants often need to continuously process materials, generate content, and revise results based on feedback. The model must not only have the general capabilities to handle these tasks, but also keep multi-turn calls sufficiently responsive. "Intelligence-efficiency ratio," a core characteristic of the Bailian model series, is continued and further strengthened at the Ling-3.1-flash model level.
In Artificial Analysis's public evaluations, 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 addition, Ling-3.1-flash's blended pricing is about $0.19 per million Tokens, placing it in the faster-output, lower-price range, balancing generation efficiency and cost advantages for high-frequency invocation scenarios.
For local deployment, the team applied MXFP4 quantization to routed experts while keeping dense layers in block-FP8, reducing weight footprint from about 534GB to 292GB. In the corresponding Hopper GPU test configuration, service deployment dropped from 8 GPUs to 4 GPUs, allowing teams with server-grade GPUs to run the model with fewer cards.
Previously, the Ling3.0 series had been practiced in open source across multiple types of devices, covering deployment paths from servers to personal computers and mobile terminals. For example, on the server side, a single NVIDIA DGX Spark can run Ling-3.0-flash to drive a coding agent to build and test a small game; on the personal computer side, a 16GB MacBook Air can run Ling-3.0-tiny as an always-on local assistant; on mobile, the Galaxy A324G validated local inference and tool routing prototypes via the Tiny GGUF version, while the Xiaomi 14T Pro achieved local conversation with the Tiny version using PocketPal.
According to reports, the Ling-3.0 series models have had nearly 1 million downloads on Hugging Face in the past 30 days, and the community has provided various adapted versions such as Flash GGUF, Tiny GGUF, and Flash MLX4bit. With the open-sourcing of Ling-3.1-flash, the Bailian model series provides developers with a more complete model foundation that can be independently verified and re-adapted, while also signifying that model capabilities are moving closer to real business delivery, further compressing the pipeline from requirement documents and code repositories to production environments.
As the model research force behind Ant Group's AI business development, Bailian provides important model capability support for consumer AI applications and enterprise-grade agents. Rather than merely pursuing model parameter scale, Bailian focuses on bringing model capabilities into real businesses to deliver results. In complex professional fields such as finance and healthcare, Bailian has successively launched enhanced models such as Ling-3.0-flash-Fin and Ling-3.0-flash-Sante based on its foundation models, further transforming general model capabilities into professional capabilities in real business processes such as investment research and health management.