Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Conductor 0.69.0 introduces OpenCode integration, allowing users to run models through the OpenCode harness. The update includes improvements such as macOS Keychain-backed storage for credentials, a fast mode toggle for …
Anthropic introduced Claude Tag, an AI that can be tagged into shared conversations, while a new paper examines stability in AI agents. The author argues that AI systems should prioritize helping users re-enter tasks gra…
Cursor (Anysphere) announced its first self-trained frontier model, Cursor foundation model, will ship within weeks. The model is not yet released, with community guesses averaging July 14, 2026. No benchmark scores or v…
XAI's Grok 5 remains unreleased as the company's Q2 2026 window closes, with crowd-sourced predictions now pointing to an August 1, 2026 release date. The probability of a June 30 launch has collapsed to approximately 10…
Anthropic has reportedly finished training Claude Mythos 5.1, the successor to its suspended Mythos 5 model, but the new model is not yet released. OutYet.ai tracks release surfaces and estimates an average release date …
Anthropic has reportedly finished training Claude Fable 5.1, the successor to the suspended Mythos 5 model, but the model is not yet released. OutYet.ai tracks release dates and shows an average guess of August 1, 2026, …
NVIDIA released Nemotron 3 on December 15, 2025, offering businesses a specialized AI model they can trust. The model series, including Ultra and Super 100B variants, achieved top benchmarks in instruction following (96)…
Ryan Tsuji, CTO at airCloset, built a knowledge graph called code-graph spanning 46 repositories using static analysis with tree-sitter to trace cross-repository dependencies. The graph addresses limitations of AI in und…
Researchers Charles Ye, Jasmine Cui, and Dylan Hadfield-Menell found that large language models suffer from 'role confusion,' mistaking the style of text for its actual content, leading to successful prompt injection att…
Z.ai's GLM-5.2 ranked as the top open-weight model and No. 3 overall on Artificial Analysis' GDPval-AA benchmark with a 1524 Elo score, placing it alongside proprietary frontier systems in multi-turn agent tasks. The mod…
Unsloth released a guide on June 18 to run Z.ai's 744-billion-parameter GLM-5.2 model on local hardware using aggressive quantization, compressing the model from 1.51 TB to as low as 217 GB. The tooling enables developer…
Radical Numerics, an AI lab spun out from Stanford, emerged from stealth with $50 million in seed funding and released Omnii, a genome language model for researchers. The company aims to decode the 'grammar of DNA' to ac…
A new Pew Research Center survey finds 49% of U.S. adults now use AI chatbots, up from 33% in 2024, with ChatGPT leading at 44% adoption. However, skepticism remains high: 40% predict AI will negatively affect society in…
OpenAI launched new security tools including the GPT-5.5-Cyber model, an updated Codex Security plugin, and the Patch the Planet open-source initiative. The company shifts from vulnerability discovery to patch automation…
Sakana AI released Fugu Ultra, a frontier-level orchestration model that rivals top AI models like Anthropic's Fable 5 and Mythos Preview. The model uses a multi-agent system to delegate subtasks to expert LLMs and is av…
Claude Code creator Boris Cherny told Meta's @Scale conference that AI loops are 'for real,' describing a shift from hand-written code to agents prompting agents that continuously improve code architecture. The approach,…
Oura Health detailed its approach to evaluating generative AI features like Oura Advisor, emphasizing clinical oversight and strict guardrails to ensure safety and trust in health-related AI responses. The company evalua…
JetSpec, a new speculative decoding method, trains a causal parallel draft head over fused hidden states from a frozen target model, enabling lossless verification of candidate trees in one forward pass. On Qwen3-8B with…
Researchers introduced JetSpec, a speculative decoding method that trains a causal parallel draft head on a frozen target model to draft entire speculative trees in one pass while preserving autoregressive factorization.…
AI in Context released a video detailing the five-day crisis in November 2023 when OpenAI's board fired CEO Sam Altman, leading to employee revolt and his reinstatement. The video explores corporate governance and accoun…