Google launches Decisions API to score up to 1,000 questions per call
Gemini Decision 1 returns confidence scores in 150-250 ms for $0.10 per 1M input tokens, with no output-token charge, Google says.
Models, from Google
The day in AI, in seven stories
Today's stories were mostly about plumbing and its limits. Google, AWS and Google Cloud shipped tools that cut cost, remove account requirements or open up code, while the biggest claims, from Signal proofs to AMD ports, arrived with caveats. Meanwhile United and Australia drew lines around what agents and model makers may do.
7 stories, about 9 minutes to read them all.
AWS releases an official local emulator for testing against its APIs
The Rust-based AWS Emulator needs no AWS account and carries no service fee, but macOS and Windows users must run it in Docker.
Coding tools, from Simon Willison
DeepMind's AlphaProof Code formally verifies key parts of Signal
Google says the Lean proofs took four weeks; Wired reports the scope was limited and no new vulnerabilities were found.
Research, from The Decoder
Anthropic says Mythos ported Claude to AMD GPUs in three days
The claim comes from an investor presentation and is unverified. Anthropic says its model wrote or ported nearly all the kernels itself.
Chips and compute, from The Decoder
Google Cloud open-sources TPU Sync for KV cache and weight transfer
Apache 2.0 library moves KV cache and weights across TPU memory and network. Google reports large TTFT cuts; builds are development-only.
Open source, from InfoQ
United bars personal AI agents from booking its flights
The airline cites privacy and federal rules; agents can still show United flights but cannot complete a purchase.
Agents, from The Verge
Australia proposes mandatory audits for frontier AI model makers
The government plans a new oversight body, with steep penalties, to enforce independent testing, published safety frameworks and immediate incident reporting.
Policy, from The Decoder
