Tag
24 articles
A German AI consortium has released Soofi S 30B-A3B, an open language model that outperforms existing competitors on both English and German benchmarks.
OpenAI staffer Vaibhav Srivastav outlines how GPT-5.6 Sol's five reasoning levels align with task complexity, recommending a gradual scaling approach for optimal performance.
NVIDIA introduces Nemotron-Labs-3-Puzzle-75B-A9B, a compressed hybrid MoE LLM delivering 2.03x server throughput, leveraging hardware-aware compression and knowledge distillation.
Learn how NVIDIA's new AI model Nemotron-Labs-3-Puzzle-75B-A9B uses compression and smart design to work faster and more efficiently than previous versions, without sacrificing quality.
MiniMax is set to release a 2.7-trillion-parameter AI model that will be open-sourced, marking a major step in China’s push to dominate the global AI landscape.
Learn how NVIDIA's new AI system Audex combines audio and text processing in one powerful model, preserving text intelligence while adding speech capabilities.
Learn how to work with mixture-of-experts (MoE) language models like Tencent's Hy3 using Hugging Face's Transformers library. This beginner-friendly tutorial teaches you to load, tokenize, and generate text with MoE models.
This article explains how language model fine-tuning and personalization work in AI systems, using the comparison between Gemini and Claude's email reply generation as a practical example.
Learn to work with NVIDIA's Nemotron-Labs-TwoTower, a hybrid language model combining autoregressive and diffusion approaches for improved text generation throughput.
This article explains the technical foundations of diffusion language models, how ByteDance's iLLaDA works, and why this new approach may challenge traditional autoregressive models.
GLM-5.2's OpenAI-compatible API offers a streamlined way to integrate advanced AI features like reasoning effort control, function calling, and long-context retrieval into applications.
Zhipu AI's GLM-5.2 closes in on closed-source leaders in coding benchmarks, trailing only Anthropic's Claude Opus 4.8 by one percentage point in the FrontierSWE test.
