18 results for “tool calling”
NVIDIA has released NemotronLabs VoiceChat 11B, an 11-billion-parameter end-to-end speech-to-speech model that performs streaming speech understanding. VoiceChat 11B gives researchers and infrastructure teams a reference implementation.
LG CNS researchers unveiled KOPA-Bench, 145 tasks over live Korean public APIs, plus EDGE, a verified data-synthesis pipeline. A 9B model tuned on its data nearly matched an untuned 27B sibling.
NVIDIA outlined an agent-eval framework pairing step-level tool-call scoring with end-to-end task completion. Charts below show vendor PinchBench accuracy and GPU-hour claims for Nemotron 3.5 Lightning.
Alibaba Cloud added Qwen-Audio 3.1 Realtime to Model Studio with tool calling, semantic turn detection, and full-duplex control. Alibaba reports background-speech responses falling to 13%, though some benchmarks slipped.
Google DeepMind released Gemini 3.8 Live with Live Avatar, bringing near real-time video personas to Gemini Enterprise. The feature supports 97 languages with lip-sync and async tool calling for smoother dialogue.
Alibaba launched Qwen3.8-Omni-Flash on September 18 with a 1M-token omnimodal window for agentic audio-video work. The model is live on Qianwen and Alibaba Cloud Model Studio APIs.
NVIDIA's NIM 2.0.12 stack boosts Nemotron 3 Ultra throughput to 1,997 tokens per second on four B200 GPUs, a 2.5x gain at 50 TPS per user. Enterprises can replay this profile using AIPerf.
NVIDIA released Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters built for high-volume agent execution. The open model cuts inference cost for always-on agents by handling routine tool calls while frontier models focus on planning.
OpenAI released the Agents API in public beta, providing a managed orchestration layer for long-running agents with hosted sandboxes and multi-agent support. This allows developers to build production agents without custom orchestration infrastructure.
Microsoft Foundry added structured outputs, web search, web fetch, MCP connector, and tool search to Azure-hosted Claude models. The update lets regulated enterprises run agentic workflows without moving data outside their Azure subscription.
MIT News profiled three researchers who moved from the MIT-IBM Computing Research Lab to IBM Research, turning PhD work on reinforcement learning, trustworthy AI, and quantum theory into production systems. Their projects address deployment bottlenecks in agent adaptability, runtime safety, and provable quantum advantage.
NVIDIA's BioNeMo Agent Toolkit now runs inside Anthropic's Claude Science, letting AI agents orchestrate protein structure prediction with MSA Search, OpenFold3, and Boltz-2 NIM microservices. The benchmark shows paired MSA input is essential — interface confidence collapses from ~0.85 to ~0.15 without evolutionary alignment.