Tool Use: The New LLM Benchmark
Evaluating LLMs now requires testing their ability to use external tools, shifting focus from pure text generation.
📂 LLM News
Latest articles in LLM News
Chain-of-Thought Prompting Boosts LLM Logic
Developers are adopting Chain-of-Thought prompting to significantly enhance the logical reasoning capabilities of large …
FaceWall Intelligence Unveils Edge AI Suite
Chinese AI firm FaceWall releases 5 key edge models and tools in a week-long open-source event, challenging global stand…
Microsoft Leak Reveals Claude Mythos Scale
A Microsoft slide leak reveals Anthropic's Claude Mythos used 6.1x10^27 FLOPs, matching Gemini 3.1 Pro and signaling a n…
Nemotron 3 Ultra: Open Moe Hybrid for Agentic Reasoning
Nemotron 3 Ultra introduces an open MoE hybrid Mamba-Transformer architecture designed to enhance agentic reasoning capa…
Fixing GPT-5.5 'Lazy' Behavior in Excel Automation
Struggling with GPT-5.5 skipping details? Learn how to stop model laziness and ensure accurate Excel data processing wit…
Google Gemma 4 12B: Edge AI Race Heats Up
Google releases Gemma 4 12B under Apache 2.0, enabling local enterprise AI workflows and shifting the focus from cloud-o…
GPT-5.5 vs Claude Opus: The Usability Gap
Users report GPT-5.5 suffers from verbose, redundant outputs compared to Claude Opus's concise precision.
GPT-5.5 vs Claude Opus: Why Developers Are Frustrated
Developers report GPT-5.5 suffers from verbose, repetitive outputs compared to Claude Opus's concise coding answers.
Gemma 4 12B: Running on 16GB VRAM Explained
Google's Gemma 4 12B claims laptop readiness. We analyze how BF16 weights fit in 16GB VRAM via quantization and memory m…
New AI Proxy Offers Free Credits & Low API Rates
Yunxuanzhang22 launches a stable AI API proxy with free trial credits. It offers competitive pricing for GPT and Claude …
Perplexity's 'Search as Code' Cuts AI Costs by 85%
Perplexity launches Search as Code, enabling AI agents to write Python search pipelines. This boosts benchmarks and slas…