Qwen

Zhejiang targets AI leadership with 582.87 billion yuan industry
Zhejiang is positioning itself as a major AI innovation hub, backed by rapid growth in its core AI industry, expanded computing infrastructure and an open-source model…

Snorkel AI study measures the efficiency of local language models
Snorkel AI has presented a study that measures how much useful language-model performance local hardware delivers for its power consumption. The Intelligence per Watt…

GlobeBench: No AI Model Solves Half of Text World Simulation Tasks
Patronus AI has released GlobeBench, a benchmark of 1,632 tasks that tests whether language models can act as text-based world simulators. According to the company…

Datadog adds Fireworks AI as a model provider for Agent Observability
Datadog Agent Observability now supports Fireworks AI as a model provider. Teams can use models hosted on Fireworks AI to run LLM-as-a-judge evaluations and to test…

NVIDIA Cosmos Framework adds agent skills and simulation recipes for physical AI
NVIDIA has published a developer guide to Cosmos, its platform of world foundation models and tools for physical AI, covering new agent skills, Cosmos 3 Nano simulation…

Baseten Serves Qwen3-ASR 7x Faster for Batch Transcription
Baseten says its optimized serving stack for Qwen3-ASR 1.7B transcribes 7x more audio per second than out-of-the-box vLLM at high concurrency on a single RTX PRO 6000.…

VocabX: Specializing an SLM's Vocabulary Cuts Inference Cost by 77%
Dharma-AI says it cut measured inference cost by 77% in a production essay grader without changing the base model, by rebuilding the tokenizer vocabulary around the cost…

Mind Lab Opens Mint Recursive Post-Training to Enterprises, From $1,000
Mind Lab, an AI research lab focused on continual learning and post-training, is making its Mint Recursive platform available to enterprises that want models trained for…

Ai2's FairStress Tests Whether LLM Fairness Holds Under Pressure
Ai2 has released FairStress, a benchmark that measures demographic fairness in language models by looking past the final choice to how readily a model abandons a…

SageMaker JumpStart adds over 100 models from 30 providers in 2026
Amazon SageMaker JumpStart onboarded more than 100 new models across 30 distinct providers between January and September 2026, according to AWS. The company also reports…

Cambridge Team Simulates a Heatwave With 100 LLM-Driven Agents
A group of PhD students based in Cambridge, the US and Hong Kong has built an agent-based simulation in which 100 LLM-driven agents live through a heatwave. Preliminary…

Gemini 4 Argon Retakes Top Spot on Roboflow's Vision Benchmark
Google's Gemini 4 Argon has taken first place on Roboflow's overall vision benchmark, according to the computer vision company. Roboflow says the model overtakes GPT-6…

Runpod adds Nix flake for Flash and Global Volumes beta for Serverless
Runpod has released a first-class Nix flake for its Flash CLI and put Global Volumes into beta for Serverless endpoints. The updates appear in the company's Q3 2026…

ASPI test: Localised Qwen model kept Chinese government positions in Chinese
Analysts at the Australian Strategic Policy Institute (ASPI) say political assumptions from a foreign base model can survive localisation. In their tests of two AI…