Tag
11 articles
Z.ai has released GLM-5.3-Flash, a 320B-parameter, 18B-active MoE model with a 1M-token context window and native multimodal capabilities. It features a 3x reduction in attention compute and 4.4x in KV cache usage compared to previous versions.
This article explains the technical advancements in SpaceXAI's Grok 4.6, focusing on its 500K token context window, xhigh reasoning level, and post-training optimization techniques.
Pokee AI has released Pokee-Isaac 28B, a 28-billion-parameter model with a 10-million token context window designed for secure, on-premises deployment. It outperforms existing models in long-context benchmarks and delivers high-speed processing.
Liquid AI has released two open-weight bidirectional encoders, LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, which maintain fast performance at 8K context on CPU hardware.
Z.ai launches GLM-5.2 with a 1-million-token context window and two thinking-effort levels, integrating with popular platforms but without initial benchmarks.
Learn how to work with large language models that support long context windows, similar to Alibaba's Qwen3.7-Max. This beginner-friendly tutorial teaches you to prepare inputs, generate responses, and optimize memory usage for handling long-horizon tasks.
Learn to implement compressed sparse attention mechanisms that enable processing one-million-token context windows, similar to DeepSeek-V4's approach.
Anthropic explains that peak-hour usage caps and increasing context lengths are behind the rapid depletion of Claude Code tokens.
Learn about Context-1, a new AI model that helps AI systems better handle large amounts of information and complex tasks by improving how they retrieve, organize, and use context.
Learn how to prepare for the next generation of language models with extended context windows by implementing token management, text chunking, and reasoning mode simulation techniques.
As language models gain the ability to process massive context windows, experts argue that selective retrieval methods like RAG remain more efficient and reliable than simply dumping all data into prompts.
