Daily AI CatchupAI Catchup — Wednesday, August 5, 2026
Wednesday, August 5, 2026Api
Artificial Analysis Exposes How API Providers Quietly Degrade Model Accuracy
API providers are silently reducing model accuracy to cut costs—and users don't know.
MoeCursor's Mixture-of-Kittens Open-Source MoE Kernel Trains AI 2.37x Faster
Cursor's new open-source GPU kernel accelerates mixture-of-experts training by 2.37x on NVIDIA's newest hardware.
ApiGoogle Cloud API Gateway Adds Serverless Multi-Model Routing for Gemini, Claude, and OpenAI
Google's API Gateway now routes requests across Gemini, Claude, and OpenAI models from a single endpoint.
On-DeviceLiquid AI's LFM2.5-2.6B Matches 9B+ Models Running Entirely On-Device
A 2.6B agentic model matches larger models while running locally on phones—no cloud required.
InfrastructureAnthropic Reportedly Signs $10B Six-Year Cloud Deal with AI Infrastructure Startup Volta
Anthropic locks in $10B in cloud capacity from Volta to power a 133-megawatt Norway data center.