GLM-5
PulseAugur coverage of GLM-5 — every cluster mentioning GLM-5 across labs, papers, and developer communities, ranked by signal.
9 day(s) with sentiment data
-
Prime Intellect releases open framework for training trillion-parameter MoE models
Prime Intellect has launched prime-rl 0.6.0, an open framework designed for training large Mixture-of-Experts (MoE) models using agentic reinforcement learning. This new system successfully trained the GLM-5 model on so…
-
Chinese AI labs release powerful open models, challenging US frontier AI
Chinese AI labs are rapidly advancing their open-weight models, with Z.ai's GLM-5.2 achieving impressive benchmark scores and a one million token context window, rivaling top closed models like Opus 4.8 and GPT-5.5 at a…
-
AI Models: Post-Training Recipes and Future Trends Explored
A new podcast episode features Nathan Lambert and Finbarr Timbers discussing recent advancements in AI model post-training techniques. The conversation covers the industry's shift towards multi-teacher on-policy distill…
-
LLM post-training recipes evolve with new distillation techniques
A review of post-training recipes for large language models highlights significant evolution in the past year. Historically, models followed a pipeline of Supervised Fine-Tuning (SFT), reward modeling, and Reinforcement…
-
LLM benchmarks miss crucial tool-use gap for agentic AI
Public LLM benchmarks often fail to reflect real-world performance, particularly for agentic systems that rely on tool use. Models excelling in static benchmarks like MMLU may perform poorly when integrated into pipelin…
-
oMLX boosts Apple Silicon LLM performance with KV cache
oMLX, an open-source LLM inference server for Apple Silicon, has demonstrated significant performance improvements, particularly in handling large models and complex workflows. Community benchmarks and local tests highl…
-
WeiboAI releases VibeThinker-3B for advanced reasoning tasks
WeiboAI has released VibeThinker-3B, a 3-billion parameter model designed for challenging reasoning tasks like mathematics, coding, and STEM. The model utilizes an optimized post-training pipeline, achieving performance…
-
New GeoNatureAgent benchmark tests LLM agents on environmental geospatial tasks
A new benchmark, GeoNatureAgent, has been released to evaluate the performance of AI agents in environmental geospatial analysis using real-world APIs. The benchmark includes 93 tasks across various categories, such as …
-
Self-Harness enables LLM agents to improve their own operational harnesses
Researchers have developed a novel method called Self-Harness, enabling LLM-based agents to autonomously improve their own operational harnesses. This iterative process involves identifying model-specific failure patter…
-
Chinese LLMs offer 80% cost savings for high-performance pipelines
A guide details how to build a cost-effective LLM pipeline by leveraging Chinese AI models, which offer competitive performance at a significantly lower price point than Western alternatives. The setup involves a unifie…
-
Coding prowess drives AI model valuations past other metrics
The valuation logic for large language models is increasingly centered on coding capabilities, with companies demonstrating superior coding performance seeing significant financial gains and market dominance. Anthropic,…
-
AI API pricing sees major cuts for inclusionAI's Ring-2.6-1T
inclusionAI has significantly reduced its pricing for the Ring-2.6-1T model, cutting both prompt and completion prices by 75%. This change offers substantial cost savings for teams utilizing this model for high-volume i…
-
New SCDBench benchmark reveals LLM struggles with smart contract decompilation
A new benchmark called SCDBench has been introduced to evaluate Large Language Models (LLMs) used for smart contract decompilation. The benchmark includes a dataset of 600 real-world Solidity contracts with paired bytec…
-
Open-Source LLMs Evolve: Attention, Multimodality, and Efficiency Gains
The open-source LLM landscape has seen significant shifts in recent months, with Sliding Window Attention becoming mainstream, enabling much larger context windows. QK-Norm is also gaining traction as a training stabili…
-
Chinese LLM APIs slash prices, DeepSeek leads with lowest cost
Chinese AI labs have significantly reduced LLM API prices in the first half of 2026, with DeepSeek, Xiaomi, and Moonshot making these cuts permanent. DeepSeek V4-Pro now offers the lowest cost per output token at $0.87 …
-
LLM memory paging uses keyword bookmarks for long conversations
A new research paper introduces cooperative memory paging, a technique designed to help Large Language Models (LLMs) manage conversations that exceed their context window. This method replaces evicted conversation segme…
-
AI Systems Automate Scientific Research, Enhancing Discovery and Verifiability
Multiple research papers introduce novel AI systems designed to automate and enhance the scientific research process. These systems, including ResearchLoop, AutoScientists, ScientistOne, AiScientist, AutoResearchClaw, a…
-
Fireworks AI: AI agent reliability, not intelligence, is key bottleneck
A new benchmark by Fireworks AI reveals that the reliability of AI model execution, not just intelligence, is a critical bottleneck for agentic AI systems. In 720 browser automation tasks, one model failed to produce va…
-
AMD MI355 cheaper than Nvidia B200 for GLM5 serving
AMD's MI355 accelerator is now 40% cheaper than Nvidia's B200 for serving on the GLM5 architecture. This cost reduction comes 14 weeks after the initial launch of GLM5, which supports both non-MTP and other configurations.
-
New research questions effectiveness of prompt-injection attacks on RAG systems
Recent research indicates that prompt-injection attacks on RAG systems may be less effective than previously thought. Studies re-evaluating these attacks in realistic RAG pipelines, which include retrieval and reranking…