Blog
Explore the latest research results and technical insights from Ant Ling Foundation Model
What It Takes to Turn a General Model into a Domain Expert
A general model can discuss a domain long before it can be trusted to work in one. Through domain continued training, task-specific post-training, tool-use optimization, expert-built evaluation, and native multimodal feedback, we are giving Ling-3.0-flash a more rigorous way to work when the task demands it.
Why We Open-Sourced Ling-3.0 tiny & flash Base Models
A Base Model is the part of the system that comes before the product. We are releasing the completed Base Models of Ling-3.0-tiny and Ling-3.0-flash, alongside key pre-training and mid-training checkpoints, so developers can shape the next step with their own data, tasks, and goals.
Open-Sourcing Ling-3.0-tiny: Agentic AI Goes Local
Ling-3.0-tiny packs 7.9B total parameters and 1.3B activated parameters into a model built for real tool use, with BF16, FP8, and INT4 weights that run from a DGX Spark to a MacBook.
Chasing the Batch-1 Latency Floor: Ling-3.0-flash at 0.78 ms TPOT on 4× B200
Most inference benchmarks reward batching, but real users experience the next token. This is how we took Ling-3.0-flash to 0.78 ms mean TPOT at batch 1 on 4× NVIDIA B200 with SGLang, by removing host-side stalls and tightening the decode critical path.
Ling-3.0-flash for AI Agents: Open-Weight Intelligence with Lower Token Costs
An open-weight 124B model with 5.1B active parameters and a 262K context window, built for fast, low-cost inference in agentic workloads — including benchmark comparisons against Qwen 3.6 27B and hardware guidance for self-hosted deployment.
Ling-3.0-flash: More Useful Work per Token
A 124B hybrid-linear MoE built for token efficiency and sustainable intelligence. AntLing's Ling-3.0-flash is a 124B-parameter Mixture-of-Experts model with only 5.1B active parameters per token, designed to increase the amount of useful work each token can carry.
Ling-2.6-flash Release: Faster Response, Stronger Execution, Higher Token Efficiency
Ant Ling has officially launched Ling-2.6-flash—a instruction model with a total parameter count of 10.4 billion and 7.4 billion activated parameters. Faced with ever-increasing token demands, Ling-2.6-flash has chosen a different technical path: rather than simply relying on longer outputs to achieve higher scores, it adopts a systematic optimization approach centered on inference efficiency, token efficiency, and agent scenario performance. While maintaining a competitive level of intelligence, this model strives to be faster, more resource-efficient, and better suited to real-world business scenarios.