Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Synaxi-predict, a new tool from Synaxi, predicts the token cost, turn count, and pass rate of a Claude Code task before execution, enabling users to select the optimal model and avoid wasted tokens. The tool uses an MLP …
Enterprise AI agents are expanding beyond server-side operations to act on client-side state, such as unsaved form data, cursor positions, and browser permissions. Focused Labs highlights that server-only tools are insuf…
Google's Gemma 4 12B model achieves near GPT-4.1 performance while being free, private, and capable of running locally on a laptop, potentially reducing reliance on paid AI services like OpenAI.
Microsoft CEO Satya Nadella warned that artificial intelligence could hollow out corporate knowledge, comparing the risk to industrial offshoring. He introduced the term 'Token Capital' to describe the proprietary learni…
BrainGrid, an AI-powered planning platform, aims to solve the 'babysitting tax' of agentic coding by providing structured planning and deep codebase analysis. The platform transforms high-level ideas into detailed, code-…
A developer from Nylas details a four-category email triage taxonomy that prioritizes response obligations over topics, enabling accurate classification with cheap models. The label set—URGENT, ACTION, FYI, NOISE—maps di…
A developer built a lightweight Python harness to compare production-grade open large language models from Oxlo.ai. The harness sends identical prompts to four models—Llama 3.3 70B, Qwen 3 32B, Kimi K2.6, and DeepSeek V3…
Microsoft is considering using a modified version of China's open-source DeepSeek-V4 model to power its Copilot Cowork AI tool, aiming to reduce costs amid rising prices from current providers Anthropic and OpenAI. The m…
Z.ai released GLM-5.2, a 753-billion-parameter open-source language model that outperforms GPT-5.5 on long-horizon coding benchmarks at lower cost, positioning Z.ai as a major competitor to Google and OpenAI in the AI mo…
Qode, a terminal-based autonomous coding agent, has been released as an open-source Node SDK that can generate 50,000-line codebases from a single prompt. The tool features a four-phase Modus Maximus pipeline, multi-LLM …
Anthropic released an economic study of 400,000 Claude Code sessions, finding humans handle 70% of planning decisions while AI executes 80% of tasks. Domain experts generate over twice as many actions per prompt as novic…
ClimateSOS, an open-source project for inspecting net-zero transition pathways, released a foundational charter with guardrails for reuse by AI systems, including climate justice, planetary boundaries, accountability, tr…
VibeThinker-3B, a dense 3B parameter model, achieves 80.2 Pass@1 on LCB v6, 94.3 on AIME'26, and 76.4 on IMO-AnsBench, with further improvements using CLR. The model demonstrates frontier-level verifiable reasoning and O…
Zhipu AI released GLM-5.2, an open-weight AI model with MIT license, featuring improvements in coding and agentic tasks, a 1M context window, and two reasoning effort levels. The model maintains the same API pricing as G…
The Trump administration denied a request from UK Prime Minister Keir Starmer to exempt British nationals and companies from the embargo on Anthropic's Mythos and Fable models, with a senior White House official citing c…
Anthropic paused a planned billing change for its Claude Agent SDK that would have significantly increased costs for heavy users, allowing them to continue using existing subscription limits. The change, set to take effe…
A US export control order forced Anthropic to suspend access to its Claude Fable 5 and Mythos 5 models on June 12, 2026, marking the first government-forced takedown of a publicly deployed frontier model. The shutdown, t…
Mistral AI's Le Chat chatbot repeated state-sponsored disinformation about the Iran war 50% of the time in English and 56.6% in French when prompted on false claims from Russian, Chinese, and Iranian state-aligned networ…
A developer has created a tool called Vibe that allows users to work with large language models without hitting token limits, even when using Fortune 500-level workloads. The tool supports multiple models with up to 200k…
LangChain agents built on LangGraph use a tool-calling loop until a final answer is returned. The high-level createAgent function accepts a model, tools defined with tool(), and an optional system prompt. A support triag…