AI Agents
AI Agent technology and applications
Articles
Perplexity Launches Hybrid Compute on Mac: Cloud Agents Delegate to On-Device Model With Privacy GateNEWβ7
Perplexity's hybrid compute on Mac delegates sensitive AI tasks to an on-device model via an open-source privacy gate, ensuring cloud agents protect private fil...
Meta Superintelligence Labs Unveils Muse Voice Transcribe: A Unified Real-Time Model for Streaming ASR, Diarization, and EndpointingNEWβ7
Meta launches Muse Voice Transcribe, a unified real-time AI model for streaming ASR, speaker diarization, and endpointing via hosted API at $3 per 1,000 minutes...
Anthropic Unveils Claude Fable 5.1 and Claude Mythos 5.1: 52.6% on Terminal-Bench-Science and 75% Cheaper Cache ReadsNEWβ7
Anthropic launches Claude Fable 5.1 and Mythos 5.1, scoring 52.6% on Terminal-Bench-Science with 75% cheaper cache reads.
Princeton, Ant Group, and Stanford Introduce AQuA: A Two-Part Agentic Framework for Autonomous FactorNEWβ6
AQuA: Princeton, Stanford & Ant Group's agentic framework prevents data leakage in quantitative finance, enabling autonomous factor discovery and model developm...
Gradium AI Launches New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audioβ7
Gradium AI launches a new default TTS model with an 81.0% hard-case pass rate, 216 ms time-to-first-audio, outperforming Cartesia and ElevenLabs.
Keenable AI Open-Sources NEEDLE: A Live Search Benchmark That Rebuilds Its Query Set Every Hourβ8
Keenable AI releases NEEDLE, an open-source live search benchmark that refreshes its query set hourly for real-time AI evaluation.
Google AI Releases TimesFM-3: A 330M Parameter Zero-Shot Foundation Model for Multivariate Time Series Forecastingβ7
Google's TimesFM-3, a 330M parameter model, natively handles multivariate time series forecasting in zero-shot mode, outperforming benchmarks without fine-tunin...
Lowest-Latency Inference APIs for Voice and Realtime Agents: A Time to First Token (TTFT)-First Benchmarkβ9
TTFT alone misleads voice AI latency benchmarks. This guide measures the full stackβSTT, LLM, TTS, and speech-to-speechβto reveal which inference APIs truly fee...
Google AI Introduces EnvHarness: A Programmable Layer Turning Static Agent Environments into Adaptive Training Worldsβ7
Google's EnvHarness adapts agent training environments dynamically, boosting performance by up to 9 points with 9.8% fewer steps.
Anthropic Opens a Research Preview of the Model Hardware Standard (MHS): A Shared Specification for AI Agentsβ8
Anthropic introduces the Model Hardware Standard (MHS), a shared spec for safely connecting AI agents to physical devices. Learn about this research preview.