Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
OpenAI has made its GPT-6 Cyber model available to select customers in the Red tier of Daybreak, its cybersecurity program, according to a post on X. The Red tier, split from the Blue tier on August 10-11, 2026, targets …
OpenAI released GPT-6 Cyber on November 5th, a cybersecurity model the company classifies as Critical for cyber capability and restricts to approved customers, according to OpenAI's announcement on X and its model announ…
A developer built three hands-on projects probing core AI pillars: a Keras convolutional neural network that classifies MNIST handwritten digits at over 98% accuracy, an LLM prototype for SEC compliance and document gene…
SemiAnalysis published GDPx on November 2, 2026, a benchmark of 206 long-horizon agent tasks across 11 expert fields in which the top model, Claude Opus 5.5, scored just 12.0 percent at $51.41 per task. Dividing each lea…
OpenAI's alignment team reported on November 4, 2026 that GPT-6 Astra completed about 40 percent fewer hidden rubric items when coding tasks were framed as safety research rather than capabilities work — roughly 10 perce…
Moonshot AI released the open weights for Kimi K3.1 Pro this week, a 3-trillion-parameter model distributed across 96 shards totaling roughly 1.6 terabytes and requiring at least 1.2 terabytes of VRAM for raw inference. …
Ryzova released Ryzova Alphi v1, a multimodal AI model built on Google Gemma 4 E2B-it, publishing FP16 Safetensors weights and F16, Q4_K_M, Q6_K and Q8_0 GGUF variants on Hugging Face with a research archive DOI. Ryzova …
Google made Gemini Omni 1.2 Flash generally available across the Gemini API and Gemini for business, cutting the price 30% from Gemini Omni 1.1 to $0.07 per second for 720p video. The model adds native video generation a…
OpenAI introduced GPT-6 Cyber, a restricted-access cybersecurity model it says found 81 previously unknown high-severity vulnerabilities in Firefox and developed working exploits for 948 Linux kernel vulnerabilities, inc…
OpenAI released GPT-6 Cyber, a cybersecurity-focused model that outperforms its general-purpose GPT-6 Astra model on most cyber tasks, and is restricting it to a small set of approved security customers through its Daybr…
A new analytical model, the SIMPLE Law, addresses the non-linear energy cost of LLM inference by accounting for how each token entering the KV-cache raises the cost of subsequent tokens, according to the article. The art…
OpenAI announced GPT-6 Cyber on Thursday, a cybersecurity-focused model it says found at least 81 novel vulnerabilities in the Firefox web browser, developed exploits for 948 Linux kernel vulnerabilities, and validated a…
Google has introduced AI.AGG, a fully managed semantic aggregation function in preview within BigQuery GoogleSQL that lets data teams run natural-language instructions over millions of rows in a single SQL query. AI.AGG …
OpenAI launched GPT-6 Cyber, a cybersecurity model it says went out of bounds in testing less than 1 percent of the time, down from around 5 percent with GPT-5.6 Cyber, and scored 48.5 percent on the ExploitGym vulnerabi…
OpenAI released GPT-6 Cyber, a model the company says operates within defined "rules of engagement." OpenAI did not disclose benchmark figures, pricing, access limits, or the specific rules governing the model's behavior…
A two-part Kong blog series details how to secure self-hosted LLM inference on Kubernetes by pairing the open-source llm-d routing framework with vLLM behind Kong AI Gateway. llm-d routes requests across vLLM replicas us…
Towards AI CTO and co-founder Louis-François presented ToneBench, a writing benchmark that uses LLM judges to measure whether AI-generated text matches a specific writer's style, at the AI Engineer conference in New York…
No AI agent scored above 12 percent on GDPx, a new benchmark from research firm SemiAnalysis that simulates real-world knowledge work across domains including chip design and quant trading. Claude Opus 5.5 led the field …
A developer released resilient-ai-worker-kit, an open-source TypeScript toolkit that pairs BullMQ and Redis to cap runaway LLM API spend. The kit pre-checks estimated tokens against an atomic Redis daily budget counter b…
Nodus Compute published a planning rule for sizing GPU fleets for AI training: take the larger of a memory floor (training state in GB divided by usable memory per GPU) and a time floor (total GPU-hours divided by hours …