AI Pulse · Frontier Labs Watch

Google just pushed out Gemini 3.7 Flash, and the timing says something on its own. It’s only been three weeks since 3.6 shipped, and this new version is priced at half the cost. Input tokens run $0.75 per million through the end of 2026, then jump back to $1.50 in January. Google is aiming this one squarely at coding and agent work, and the numbers back that up. On the DeepSWE benchmark, which tests how well a model handles real software engineering tasks, the score went from 49.0% to 65.3%. That’s a big jump for a model update, not a small tune-up.

It’s already live everywhere that matters: the API, AI Studio, Antigravity, and OpenRouter, all on the same day it was announced. No waitlist, no staged rollout. If you’re using any Google model in your stack right now, this one is available to test today, as Google detailed in its announcement.

For an owner who isn’t deep in AI infrastructure, here’s the practical read. If you’re running or planning to run coding agents, whether that’s for internal tools, automation, or customer-facing features, this model is now cheaper and better at exactly that work than its predecessor was three weeks ago. That pace of change is the real story. Whatever you build today on a specific model may be outclassed, and cheaper, within a month. This is exactly the kind of shift that’s easy to miss if you’re not tracking model releases closely, which is part of why we build model comparisons and cost tracking into an AI command centre, so you can see when a swap actually pays off instead of guessing.