Ling 3.0 Flash is now available on AI Gateway

Ling 3.0 Flash from Ant Group is now available on AI Gateway. The model is free to use for the next three weeks, through August 3rd. That window is a short, concrete period to trial the model on production-shaped traffic without adding a line item.

Ling 3.0 Flash is a Mixture-of-Experts model with 124B total parameters and about 5.1B active per token. It exposes a 256K token context window and can run in thinking and non-thinking modes. The sparse activation profile keeps compute closer to a mid-size dense model while the full parameter budget stays available for capacity across tasks.

For engineers building agent loops, the design targets token-efficient agentic inference at production scale. Lower active parameters per token cut cost and latency on multi-step tool use, and the long context reduces how often you have to truncate or re-summarize long tool traces and session history. Thinking versus non-thinking modes let you spend more tokens only when a step needs deliberation.

Putting a free Ant Group MoE on AI Gateway sits the model next to whatever else you already route through that layer. Teams can compare quality, latency, and token use on the same paths as their current providers without a separate integration for a three-week eval. Sparse large models with dual inference modes are the pattern many builders already plan around for agent workloads.

Use the free window to measure end-to-end agent cost and quality on real traces, not synthetic chat. Compare thinking mode on hard planning steps against non-thinking mode on routine tool calls, and watch whether the 5.1B active-parameter path holds quality at the token budgets you actually ship. After August 3rd, re-check pricing and default routing before locking this model into production traffic.