More posts
In my opinion, 2026 is year zero for general-purpose agents. once everyone's running one, subscription-based skills become a real revenue model. saas today is built for humans. the next wave is saas wrapped as an api skill that agents subscribe to on their own.
But I've been running a general agent on vision stuff lately with M3, and it's surprisingly good. Other multimodal models are either expensive or bad. The cheaper ones aren't even native multimodal. M3 is the only one that gets all three right: price, quality, and native multimodal.
. 1.6T params, ~48B active, native multimodal, aimed at long-horizon agents, trained start to finish on Chinese hardware. Images get quantized into multi-level discrete tokens, and that's where Visual Coding and 3D pick up. On the RL side, failed trajectories get turned back into verifiable training tasks. W4A4 inference, 43% faster prefill. Meituan's own evals put it between Claude Opus 4.8 and Opus 5.
It stacks layers instead of parameters. The decoder has 32 layers, but layers 4 through 27 run twice with shared weights, so a forward pass does 56 layers of work while the checkpoint stays at 9B. Their own eval numbers: 71.38 on BIRD-Dev and 13.72 on AutomationBench, both above Qwen3.5-27B. Terminal Bench is still behind.
