Per-token prices have plummeted, but actual AI costs have risen instead due to a surge in usage driven by agent architectures.
Xiaomi has unveiled Xiaomi-Robotics-1, a robot policy model built on 100,000 hours of embodiment-agnostic pretraining.
NVIDIA has unveiled Cosmos 3 Edge, a 4B-parameter open world model for edge devices.
Experiments demonstrate that a well-designed LLM harness (RLM) can generalize from training on short tasks alone to tasks 8-32x longer and to new domains.
Ramp cut AI costs by over 25% and reduced error rates using dynamic LLM routing based on online learning.
Kimi K3 is the strongest open model available today, but it still trails the closed-model frontier by several months.
Google is developing a 'Frozen v2' server chip that could be up to 10 times more efficient than existing AI chips, targeting a 2028 release.
AMD will ship its first rack-scale AI system, Helios, this year, with Microsoft, Meta, OpenAI, Oracle, and others confirming adoption.
Moonshot AI has launched Kimi Work, a desktop AI agent that supports local file integration, web automation, and 24-hour scheduling.
AI Gateway has introduced a Service Tiers feature that lets users optimize latency, throughput, and cost according to their use case.
Vercel precompiles bytecode when building Python functions, significantly reducing cold start time.
OpenAI has brought on Nubank founder David Vélez and BNY CEO Robin Vince as board members.
Vercel Connect has added preset connectors for over 90 services including Shopify and Jira.
Vercel MCP now supports payment functions for Vercel products, including team plan upgrades and domain purchases.
Poolside's open-weight model Laguna S 2.1 has been added to AI Gateway.
xAI has launched a Grok add-in for Outlook that supports email summarization, reply drafting, and inbox management.
Delivery confirmation information and proper submission timing decisively increase win rates in 'product not received' disputes.
CalibAtt is a technique that speeds up video generation models by up to 1.58x by identifying unnecessary connections during attention computation.
Gemini 3.6 Flash and 3.5 Flash-Lite models have been added to Vercel AI Gateway.
This proposes a technique that generates training data for agents using only API specifications, without a real execution environment.
An open-source system called Grabette has been released that lets anyone easily collect manipulation data and build training datasets without needing a robot.
AI email assistant Fyxer adopted ElevenLabs' Scribe v2, improving speaker identification accuracy in meeting notes and boosting user conversion rates.
ElevenLabs is opening its first office in Toronto and doubling the size of its local team to expand into the Canadian market.
The PyMuPDF4LLM 1.28 update has significantly improved the accuracy of extracting complex table structures from PDFs.
AI Briefing