Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Vercel's engineering team found that deleting 80% of its AI sales agent's tools significantly improved performance, challenging the assumption that more capabilities lead to better results. The subtraction principle sugg…
Chinese AI labs are using gray market access to Western AI models for distillation attacks, training their own models on outputs from US providers like OpenAI. This practice, which involves querying APIs at scale through…
Z.AI, formerly Zhipu AI, released GLM 5.2, an open-weight language model that outperforms GPT 5.5 on design benchmarks. The model features multi-token prediction and offers API access and self-hosting options at lower co…
Zhipu AI released GLM-5.2, an open-weights model under MIT license that scores 74.4% on FrontierSWE, beating OpenAI's GPT-5.5 and trailing Anthropic's Claude Opus 4.8 by one point at roughly a sixth of the cost. The mode…
Hugging Face researchers introduced a new benchmarking methodology for open-source coding agents, measuring not just accuracy but also efficiency across different models and library revisions. Using the transformers libr…
Elastic 9.4 introduces two tools for building custom integrations: Automatic Import, which now supports multiple data streams and increased log sampling, and Integration Skills, an open, LLM-agnostic agentic toolkit for …
Noam Shazeer, co-inventor of the transformer architecture and co-lead of Google's Gemini program, is leaving Google to join OpenAI, despite Google paying approximately $2.7 billion to retain him less than two years ago. …
A research team from CUHK Shenzhen and Tencent's Hunyuan released GameCraft-Bench, a benchmark testing AI coding agents on building complete, playable games in the Godot engine. The top agent, Claude Code on Anthropic's …
Chinese lab Zhipu's commercial arm Z.AI released GLM-5.2, a 753-billion-parameter open-source language model with a 1-million-token context window, on June 17. The model achieves near-frontier performance on agentic codi…
Microsoft announced at Build 2026 that VS Code now supports bring your own language model key (BYOK), allowing developers to use models from providers like Azure, Anthropic, OpenAI, or local models via Ollama directly in…
Hugging Face's PEFT library challenges the dominance of LoRA in parameter-efficient fine-tuning, revealing that while LoRA accounts for over 98% of PEFT usage on the Hugging Face Hub, alternative methods may offer better…
Z.ai released GLM-5.2, an open-weights AI model under an MIT license, positioning it between Claude Opus 4.7 and 4.8 in performance while costing less than a fifth of Opus on output tokens. The model features a 1M-token …
Claude Code users increasingly output HTML instead of Markdown to leverage richer browser affordances for plans, specs, and reports, but this workaround reveals a missing artifact layer in the product. The practice compr…
On June 18, 2026, open-weight model GLM-5.2 scored 74.4% on FrontierSWE, trailing Claude Opus 4.8 by one point, signaling the collapse of the scale myth as smaller models rival frontier performance. The agent stack is re…
Mistral CEO confirmed a new open-weight Mixture-of-Experts (MoE) model family, with early access expected in July 2026. The model is not yet released, and community guesses average a release date of July 15, 2026.
Z.ai released GLM-5.2, an open-weight 744-billion-parameter coding model with a 1-million-token context window, on June 26, 2026. The model is available via API, though no benchmark scores have been published yet.
Ampcode's Librarian AI agent is now approximately three times faster and 43% cheaper after switching to OpenAI's GPT-5.5 model with websocket mode, reducing average search latency from 237 seconds to 81 seconds and cost …
A comparative analysis of scale-out RoCE network traffic patterns in training large language models reveals that hyperscale deployments like xAI's Grok 4.0 generate network loads orders of magnitude higher than smaller-s…
A new approach to agent safety, called deterministic agent safety, separates the LLM's reasoning from its actions by placing the model in a sandbox with no credentials or network access, and exposing only predefined acti…
Chinese AI lab Z.ai released GLM-5.2, a 753B-parameter open-weights text-only LLM with a 1 million token context window, under an MIT license. The model leads the Artificial Analysis Intelligence Index among open-weights…