How Google’s New Gemini Rates Work and How to Track Your Usage
THE SO WHAT
Google moving to more granular Gemini quotas is a reminder that AI usage is metered, not magic—your bill follows tokens and tool calls, not “number of chats.” If you’re building on Gemini, wire usage tracking into your product now or you’ll discover margin problems only when invoices land.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIGPT-5.6 used a prompt to close a 30-year gap in convex optimization
If a prompt to GPT-5.6 can surface a path on a 30-year-open convex optimization problem, the frontier is shifting from “can the model do it” to “can humans ask the right question and verify the answer.” Research-heavy orgs should be piloting LLMs as first-pass theorem and proof-search engines, with human experts focused on validation and refinement.
Applied AIAlibaba open-sources its AI chip software stack at WAIC, targeting Nvidia’s CUDA lock-in
Alibaba T-Head open-sourcing SAIL for its Zhenwu AI chips is a direct attempt to erode CUDA’s moat by lowering migration costs at the software layer. If you’re planning large-scale training or inference in China or with Chinese partners, you now have to model a multi-stack world where CUDA is no longer the only default.
Applied AIFable 5 vs. GPT-5.6 Sol on an NP-Hard Problem: Does /goal help?
Head-to-head tests of Fable 5 versus GPT-5.6 Sol on NP-hard problems with and without a /goal directive highlight that prompt structure and task framing can rival raw model size. Teams building agentic systems should be investing as much in goal-specification schemas and orchestration as in chasing the latest base model.
Applied AIMoonshot’s Kimi Upends Conventional Wisdom on US Lead Over China
If Kimi is competitive with top US models, the assumption of a durable US lead in frontier AI capability is weaker than many Western roadmaps assume. Multinationals operating in or with China need to treat local model parity as a near-term planning input, not a distant scenario.