#LLM
Curated coverage on «LLM»: releases, research, benchmarks, open-source.
OpenAI brings GPT, Codex and Managed Agents to AWS for enterprise AI development
OpenAI has added GPT models, Codex and Managed Agents to AWS so companies can run and scale AI tools inside Amazon's own secured cloud environment.
Local vision for z.ai GLM-5.1: an 8B model closes 70% of the gap to the frontier
A developer gave a cheap z.ai GLM-5.1 coding agent local "vision" via an MCP server and qwen3-vl:8b, so it can check its own UI from screenshots.
Aurora proposes a manifesto for durable AI, built on memory, identity and growth
A manifesto on "durable AI" has appeared on Habr AI. Its author argues systems should be judged not by benchmarks but by whether they keep their identity, change…
Claude Can't Save a Bad Brief: Why AI Delivers Convincing but Weak Results
Proton CEO: Privacy Is Possible in the AI Era, but Autonomous Agents Bring a New Risk
ChatGPT growth slows as app deletions climb, raising risks for OpenAI's IPO
Google reports record Search queries and 19% revenue growth
OpenAI explains where GPT-5's "goblins" came from: how a personality bug leaked into the model
OpenAI says GPT-5.1, GPT-5.4 and GPT-5.5 started talking about goblins and gremlins because of Nerdy-mode training, whose style spread to the whole model.
OpenAI publishes a cyber defense plan for the AI era, aimed at governments and critical infrastructure
OpenAI's five-part plan would widen defenders' access to powerful AI tools, tighten oversight of how they are used, and add new security features to ChatGPT.
OpenAI explains how ChatGPT detects threats and blocks dangerous violence scenarios
Marcin Moskała reviews GeminiAI: what the audit revealed about coroutines and Android architecture
Anthropic: Leaked Claude Code Source Reveals a Sophisticated Agent Architecture
Saiga Llama 3 8B on 10 GB of VRAM: How Habr Reached 93% Accuracy on War and Peace
Veai shows how it tests an AI agent inside JetBrains IDEs without depending on the model
Veai described its UI automation approach for a JetBrains IDE plugin: the team separated the deterministic interface from LLM responses and cut false test failures.
Qwen 3.6 Plus Beats DeepSeek V4 Pro in Russian Battle Test and Costs Less
In a fresh comparison of six April LLMs, Qwen 3.6 Plus scored 92 on Russian content, beating the new DeepSeek V4 Pro, which scored lower and costs noticeably more.