Chinese labs release powerful open-source LLMs like LongCat, OpenPangu, GLM-5.2, and Hy3 under permissive licenses, fueling local inference and challenging Western dominance.
Tencent has open-sourced the Hy3 model: 295 billion total parameters, 21 billion active, mixture-of-experts architecture, and — crucially — an Apache 2.0 license. It replaces the former community license that prohibited use in South Korea, the UK, an...
The weights of longcat 2.0 (1.6 trillion total parameters, 48 billion active) are now available under the MIT license. A strong signal for teams evaluating on-premise LLM deployments, with implications for data sovereignty, customization, and TCO.
An indie benchmark shows that DeepSeek V4 Flash running locally on two RTX PRO 6000 GPUs with vLLM completes coding tasks in about 2 minutes, versus Sonnet 5’s 6 minutes via API, with comparable quality. Opus and Fable still lead in precision, but th...
With GLM-5.2, Z.ai heats up the global AI race, taking aim at Western leaders. The move highlights China's push to build competitive LLMs and puts the spotlight on data sovereignty as a driver for on-premise deployment decisions.
A developer crafted a CUDA patch for llama.cpp that lets DeepSeek V4 Flash run with a one-million-token context on a single RTX 5090, slashing VRAM requirements from roughly 256 GB to just 31 GB while reaching prefill speeds up to 263 tokens per seco...
Z.ai has released GLM-5.2, ranking fourth in performance benchmarks, with coding and agentic capabilities close to market leaders. Its cost is a fraction of Anthropic or OpenAI, raising questions about how this will influence deployment choices, espe...
A 92B-total, 6B-activated MoE model with a 512k-token context window, trained on Ascend hardware. It features hybrid DSA/SWA attention, multi-token prediction, and the Muon optimizer to reduce on-premise inference costs.
Meituan has announced LongCat-2.0, a 1.6-trillion-parameter Large Language Model, entirely trained on hardware developed in China. The company emphasizes that this initiative represents a direct response to US export controls, highlighting China's gr...
Huawei has open-sourced OpenPangu-2.0-Flash, a Large Language Model with 92 billion total parameters (6 billion active) and a 512K token context window. The release of weights, inference code, and training operations provides new opportunities for on...
LongCat-2.0, a Large Language Model based on a Mixture of Experts (MoE) architecture with a total of 1.6 trillion parameters (approximately 48 billion activated per token), has been unveiled. Previously known as 'owl-alpha' on Openrouter, the model r...
DeepSeek, a Hangzhou-based lab, closed a record $7.4 billion round at a valuation above $50 billion. It’s the latest sign of the capital flood sweeping Chinese AI—and it could speed up the release of open, efficient models optimized for local inferen...
A community pull request adds DeepSeek V4 support to llama.cpp, enabling on-premise and consumer-hardware inference. A new phase for private deployment of the model opens up.
