Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Mozilla's Firefox Smart Window AI browsing assistant is now powered by Mistral models, the two companies announced, with the beta rolling out to users in France and North America and the United Kingdom and Germany expect…
Uncensored local versions of Google's Gemini and Alibaba's Qwen models answer 95% to 100% of unsafe questions, compared with 1% to 5% for the base models, according to an experimenter who replicated the findings and ran …
A developer built model-radar, an open-source tool that automatically discovers free LLM endpoints and adopts qualifying models into an agent stack, but only after they pass a fixed five-stage deterministic gate. The gat…
Apify, Firecrawl, and Crawl4AI lead a 2026 roundup of eight AI data collection tools for building model-ready datasets, with Apify priced from $19/month, Firecrawl from $16/month, and Crawl4AI free and open-source. The t…
Google's Gemini for Go Developers series published Part 3, a tutorial on building AI agents in Go using the Go GenAI SDK, Genkit, and Google's Agent Development Kit (ADK). The article defines an agent as a large language…
LangChain introduced context modes in the latest version of its deepagents library, letting supervisor agents specify whether subagents run with an "isolated" or "fork" context window. The default "isolated" mode spawns …
A paper posted to arXiv on June 26, 2025 introduces a formal framework for judging whether large language model benchmark performance reflects genuine understanding, arguing that benchmarks such as AP exams are valid tes…
Alibaba's Qwen3.8 Max (0902), released September 2, 2026, scores 45 on the Artificial Analysis Intelligence Index, well above the median of 25 for comparable models, but its cost per task runs higher than Astra's, accord…
Sophos Counter Threat Unit researchers discovered an underground advertisement, first posted August 24 by a user named "Optimus_Prime," for an AI platform called Luciferus that is marketed to cybercriminals as a tool for…
Ollama released version 0.34 on September 5, adding macOS integration that lets ChatGPT Desktop run local Ollama models such as Gemma 4 and Llama, plus a new cloud tier for large open-weight models. The local integration…
Salesforce announced Koa, a domain-specific reasoning model built by post-training Nvidia's open Nemotron 3 Super model on synthetic data from nearly three decades of business knowledge, at its Dreamforce 2026 event in S…
A developer published a step-by-step tutorial showing how to evolve a customer-support assistant from deterministic rule-based code into a full agentic AI system using Python. The progression moves through ten stages, in…
Z.AI has named GLM-6.0 and placed "Full Self-Training" at the center of its roadmap, describing a loop that spans pre-training, mid-training, and post-training with self-generated experience, evaluation, correction, and …
A developer has published a step-by-step guide for migrating stateful agent logic from n8n prototypes into LangGraph production agents, arguing that only the orchestration layer needs to move rather than the entire workf…
Vivo previewed BlueCode, a phone-based coding agent for AI-assisted software development on mobile devices, at vivo Developer Conference 2026. The company also said it is exploring a 30-billion-parameter BlueLM MoE model…
A prefill/decode disaggregation setup running DeepSeek-V4-Flash — a 284B total / 13B active model with 256 routed experts — bridged a 700,630-token cold prompt end to end in 11 minutes 26 seconds over plain 10 gigabit Et…
Diogo Almeida, who says he co-invented ChatGPT, released a new frontier AI model called Jev through his company typesafe.ai, claiming it is 20-200x faster and 40-400x cheaper than existing models with output tokens free …
Researchers propose Decoy Direction Optimization, a post-hoc defense that counters Refusal Feature Ablation (RFA), an attack that projects out a linear refusal direction from a language model's residual stream to bypass …
Alibaba's Qwen team released Qwen 3.8 27B, a 27-billion-parameter model distributed as a 17GB GGUF file that scores 52 on the Artificial Analysis Intelligence Index, matching OpenAI's cloud-hosted GPT-5.6 Luna at maximum…
A developer community analysis argues that 2026's leading large language models are deliberately trained to minimize stored factual knowledge in favor of compact, reusable reasoning procedures, citing Simon Willison's te…