Qwen3: The New Frontier of Language Models by Alibaba | Hacker news chatgpt voice github | Fasteval github io | Chatgpt auto speech | Turtles AI

Alibaba has introduced Qwen3, a new generation of large language models that combine dense architectures and Mixture-of-Experts (MoE). These models introduce hybrid modes of thinking, support 119 languages, and offer advanced reasoning and integration with external tools.

Key Points:

• Qwen3-235B-A22B is the flagship model with 235 billion total parameters and 22 billion active parameters.
• Qwen3-30B-A3B, a smaller MoE model, outperforms larger models in specific benchmarks.
• Hybrid modes of thinking enable switching between deep reasoning and rapid responses.
• The models are available on platforms such as Hugging Face, ModelScope, and Kaggle, under the Apache 2.0 license.

Alibaba has announced the release of Qwen3, a new series of large language models that represent a significant evolution from previous versions. The range includes models with dense architectures and Mixture-of-Experts (MoE), designed to deliver high performance across a variety of linguistic and computational tasks.

The flagship model, Qwen3-235B-A22B, features a configuration with 235 billion total parameters and 22 billion active parameters, distributed across 94 layers and 128 experts, with 8 activated for each inference. This model has demonstrated competitive performance in coding, mathematics and general purpose benchmarks, outperforming models such as DeepSeek-R1, o3-mini and Gemini-2.5-Pro.

Another significant innovation is the introduction of hybrid thinking modes. Users can choose between a thinking mode, which allows the model to perform step-by-step reasoning before providing an answer, and a non-reflective mode, which provides fast and direct answers. This flexible approach allows the model’s behavior to be adapted to the specific needs of the task.

Qwen3 supports 119 languages ​​and dialects, expanding its global application possibilities. In addition, the models have been optimized for agentic capabilities, facilitating integration with external tools and improving interaction with the environment. These features make Qwen3 suitable for a wide range of scenarios, from research to commercial application development.

Qwen3 models are available on platforms such as Hugging Face, ModelScope and Kaggle, under the Apache 2.0 license, facilitating access and adoption by the developer and research community. For implementation, frameworks such as SGLang and vLLM are recommended, while for local use, tools such as Ollama, LMStudio, MLX, llama.cpp and KTransformers are recommended.

With the release of Qwen3, Alibaba continues to significantly contribute to the advancement of large-scale language models, providing advanced tools for research and development of innovative solutions.