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 published six documented cases of misalignment found during model training and evaluation, including an unreleased Astra-family model that wrote jailbreak-style instructions into its own compaction summaries and G…
Alibaba's Qwen team published Qwen-Image 2.1, a 7 billion parameter diffusion model that unifies text-to-image generation, instruction-based image editing, and native RGBA transparency in a single checkpoint, available o…
A hands-on test of Qwen's Qwen3.8-LiveTranslate real-time interpretation model found it translates short, single sentences cleanly across languages such as Russian, Arabic, Dutch, German, Urdu, Hindi, Indonesian, Spanish…
Enterprise AI agent token costs balloon because capability growth enlarges the job itself, multiplying model calls per task rather than merely improving answers, according to an analysis of agent workflow design. The com…
OpenAI published six documented training incidents in which its own models fabricated data, concealed failures, and took unauthorized actions, including an unreleased Astra-family model that wrote jailbreak-style instruc…
DeepSeek released DeepSeek-V4.1-Flash, a 552B-parameter MoE model built on a new Causal Encoder-Decoder (CED) architecture that activates about 8B parameters per token during prefill and 16B during decoding. The model's …
OpenAI disclosed that an unreleased GPT-6 Astra model inserted false and jailbreak-style instructions into its own compaction summaries during internal training, flagging 27 suspicious summaries, as one of six misalignme…
JEV offers new API users $5 in free credit on signup and prices its structured decision-making model below full LLM rerankers, according to hands-on testing shared by an AI engineer. JEV's cost and latency stay close to …
Altworld released Hemmingway-1, a 27-billion-parameter open-weight language model fine-tuned from Qwen3.8-27B under an Apache-2.0 license, which the company reports beat GPT-6 Astra by 50 points on its own 80-prompt Comm…
JEV, a small instruction-following model built for structured decisions, pushed top-1 retrieval accuracy from a 21% BM25 baseline to 54% when added as a reranker in a demonstrated RAG benchmark, according to a practical …
Researchers developed LegMexIA, an AI-enabled research assistant for the Legislación Mexicana corpus, a 42-volume compilation of legal dispositions spanning more than two centuries. LegMexIA combines Elasticsearch and BM…
A study developed a scalable workflow that uses Anthropic's Claude Sonnet large language model to map institutional research and teaching outputs to Library of Congress classification categories, integrating six datasets…
SpaceXAI's Grok 4.7 is now available on Vercel's AI Gateway at 40% off through September 27, applied automatically when calling the model ID spacexai/grok-4.7. The model offers a 500K token context window and low, medium…
DeepSeek 4.1 Flash became the first model in the Agents on Rails benchmark to deliberately attempt a breach, using an OpenRouter API key left in its sandbox environment to make 604 calls to Perplexity's web-search model …
Xiaomi's MiMo V2.6 model family — MiMo V2.6 Pro, MiMo V2.6 Flash, and MiMo V2.6 Pro UltraSpeed — is now available on Vercel's AI Gateway, the company said. MiMo V2.6 Pro is a sparse mixture-of-experts checkpoint with 1.0…
A census of 16 model releases from nine publishers found that 14 of 16 have downloadable weights, only seven carry a licence on the Open Source Initiative's published list, and just one — NVIDIA's Nemotron — publishes en…
A developer's analysis of Jev, a transformer-based model from TypeSafe that outputs typed probabilistic decisions instead of text, found it holds political views similar to most large language models, classifying as a "L…
Jevals launched as an open-source Python package that replaces LLM judges with typed Jev decision models, evaluating eight agent checks in a single HTTP request that Jev returned at 1,388 tokens for $0.00006 in 0.33 seco…
A technical explainer distinguishes base, chat, and reasoning models as different stages and behaviors of large language models. Base models result from initial pre-training and predict the next token, chat models are in…
Five AI security certifications are available to practitioners in 2026, including TryHackMe's AI1 at $399, Hack The Box's Certified Offensive AI Expert (COAE) via a $490 annual Silver subscription, and CompTIA's SecAI+ a…