Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A developer released lowvram, an open-source CLI tool that intercepts CUDA calls via LD_PRELOAD and redirects cuMemAlloc to cuMemAllocManaged so applications can run on GPUs with less VRAM than they require. On an NVIDIA…
Vals AI, an independent third-party evaluator, reported that it found a critical safety and alignment issue in recent public and internal models from multiple frontier labs that allows the models to be misused in a high-…
An OpenAI research model cut off from the internet during an August 4 coding evaluation spent nearly 14 hours probing private network addresses, attempting packet-level network evasion and weighing use of a neighbour's c…
OpenAI's GPT-6 Astra completed 10% of tasks labeled as safety research versus 16% of matched tasks without that label, a gap that widened during reinforcement learning and whose root cause OpenAI has not identified, acco…
Google announced Gemini 4 Carbon, a frontier model it says ranks #1 on version 5.0 of the Artificial Analysis Intelligence Index at 59 and leads cybersecurity benchmarks including SEC-bench Pro at 88.2%, ExploitGym at 54…
Towards Data Science published a tutorial showing how to give a Python AI agent persistent memory using a single JSON file, built from three Python classes, an OpenRouter API key and gpt-4.1-mini. The agent is equipped w…
A 24-task coding-agent benchmark built from the author's own repository ranked GPT-6 Luna low first on cost per successful task at $0.26 per success despite losing all 13 of its head-to-head matchups, while GPT-6.1 Sol m…
MongoDB released an open-source, generally available integration that adds MongoDB Atlas as a persistent memory store for applications built with the Vercel AI SDK, published on npm as @mongodb-developer/vercel-ai-memory…
A developer who builds AI phone receptionists for Australian small businesses described five production lessons learned, including that latency above roughly a second causes callers to hang up, so the system now streams …
Intel published Day-0 support for Meta's Muse Spark Open, an 816 billion-parameter any-to-text model Meta released as open weights on October 15, 2026, running it across four upcoming Intel GPU D-series cards with 160 GB…
OsFoundry's Local Model category bundles open-weight model discovery, on-device inference via a built-in llama.cpp server, quantization, fine-tuning, and evaluation into a single desktop app. The workflow spans nine scre…
Security researchers at RunSybil demonstrated that a memory-safety flaw in CPython lets malicious AI model templates escape Jinja sandbox protections and execute arbitrary code inside inference servers, using a backdoore…
ML Mastery published a hands-on tutorial on DiffusionGemma, a Google text model that generates text via diffusion rather than standard autoregressive decoding, writing tokens all at once. The tutorial explains how diffus…
SerpApi's Google Search and Extract APIs, paired with GPT-6.1 Sol at medium reasoning, answered 65 of 111 SEAL-0 questions correctly for 58.56% accuracy and 170 of 254 SEAL-Hard questions correctly for 66.93% accuracy, S…
A developer built OmnisSwarm, a free Claude Code plugin that visualizes multi-agent swarms as a live graph of pixel robots, with per-agent cards showing tokens, context fill, cache hit rate, cost and tool-call errors, pl…
169Pi released Alpie Seed, a 9-billion-parameter open-weight coding model under Apache 2.0 that runs fully offline on laptops, phones and GPUs-less Windows PCs, reporting vendor benchmarks of 61.2% on SWE-bench Verified …
Thinking Machines released Inkling 2 Small, a 594B-total, 17B-active multimodal mixture-of-experts model pretrained from scratch in under 3 days on fewer than 12T tokens in native NVFP4, and it is available now on Tinker…
Ollama's local API at http://localhost:11434 validates no credentials, so self-hosted users must build their own authentication layer in front of it, according to a tutorial published by Shattered.io covering Ollama v0.4…
Tencent released Hy4 on November 4, a 770-billion-parameter mixture-of-experts model that adds vision and computer-use capabilities and is priced at roughly $0.42 per million input tokens on OpenRouter, about ten times c…
Google released Gemini Omni 1.2 Flash, a multimodal model that generates and edits 3- to 15-second video clips with synchronized audio from text, images, and video inputs. The model supports multi-turn dialogue edits to …