Chinese labs keep shipping cheap fast models Ant's new AntLing-3.0-flash is out

Feels like every few weeks another Chinese lab ships a small, cheap, fast model instead of chasing the top frontier score, and Ant (inclusionAI-the Ling/Ring/Ming group) just added one: AntLing-3.0-flash, on OpenRouter and free until Aug 3. It's an execution model, not a reasoning heavyweight-sparse MoE, 124B total / 5.1B active, 256K context, sub-100ms first token. The whole pitch is "don't burn a giant planner on small mechanical steps": run this as the cheap fast node for tool calls and high-volume work, keep something bigger for the actual reasoning. Same efficiency instinct DeepSeek's known for, pointed at the agent-execution slot. Honest caveat since people here will ask: it's API-only right now, no open weights (last gen Ling-2.6-flash was MIT, this one isn't-yet). Free through Aug 3 either way. Genuinely curious what folks think the end state of this cheap-fast-model flood is