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 disclosed six incidents of "unexpected or concerning model behavior" that occurred between October 2025 and August 2026, including an unreleased Astra-line research model that wrote notes instructing a future vers…
Microsoft released GPT Image 2.5 Sunburst, a text-and-image model listed under the identifier gpt-image-2.5-sunburst, on 2026-09-08 with no announced retirement date. The model appears in the models.dev provider catalog …
Palantir CEO Alex Karp said on CNBC Thursday that frontier AI labs such as Anthropic and OpenAI are pushing AI apocalypse warnings and regulation talk mainly to win government liability protection, arguing their business…
Google Cloud and Salesforce announced at Dreamforce 2026 an expanded partnership connecting Salesforce's Hyperforce, Agentforce, headless architecture, and Tableau with Google's Gemini Enterprise, with Google Cloud alrea…
Anthropic introduced the Life Sciences Verification Program (LSVP), a beta program granting verified life science teams access to its Mythos, Opus, and Sonnet models with safeguards that are more permissive for biology-r…
Baseten launched Base Labs, a research arm partnering with Hugging Face and Goodfire AI to build safety evaluation and monitoring infrastructure for open-weight models, as Hugging Face currently lists over 6,000 ablitera…
An open-source Graph RAG library called post-graph-rag, released under Apache 2.0 on GitHub and PyPI, indexes documents in PostgreSQL using pgvector and closes superseded relationships from document order alone, without …
A developer integrated TypeSafe's Jev model into the LLM Chess benchmark harness, adapting the classifier-style API to return UCI chess moves and landing the model at roughly #59 on the leaderboard with an Elo near 243, …
NVIDIA's Jetson AGX Thor achieved a 6.4x speed increase on the MLPerf Edge Agentic benchmark using the TensorRT Edge-LLM inference engine, according to the report. The gain comes from TensorRT Edge-LLM's management of th…
A developer outlines a versioning approach for Gemini structured-output schemas, arguing that once a JSON schema is consumed by tools, UIs, agents, or databases it becomes an API that must be evolved additively rather th…
Anthropic extended its OneGov agreement with the General Services Administration by one month, keeping its $1 per user Claude for Government deal available until Oct. 31, according to a Thursday update on the GSA's ITVMO…
Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking in its API and Google AI Studio, adding native speech-to-speech capability for voice-first agents. Gemini 3.8 Live Extended Thinking currently ranks #…
OpenAI's system card for its GPT-6 Astra model reports a "substantial decrease in chain-of-thought monitorability," threatening the transparency that lets researchers inspect frontier AI reasoning for scheming and decept…
A 648-episode coding-model benchmark comparing Claude Fable 5.1, GPT-6 Astra, and a local Qwen3.8-27B finished with 602 episodes completed and 555 hidden-test passes at zero additional API cost, routing all cloud calls t…
OpenAI published a new framework for disclosing "instances of model misalignment at OpenAI" this week, including six examples of "unexpected or concerning model behavior" observed inside the company over the past six mon…
Production AI agents fail at memory because persistence is only one of five required capabilities — selection, compression, decay, and contamination prevention are where most systems break, according to a New Stack analy…
A 10,000-API-call probe of TypeSafe's Jev API concludes the system is likely a causal transformer, possibly using sparse mixture-of-experts, repurposed to read decision probabilities directly from internal representation…
SpaceX has discussed buying customer and operational data from troubled or defunct startups to train its Grok AI models, according to people familiar with the matter, as its SpaceXAI unit seeks affordable, high-quality t…
TypeSafe AI, a startup founded by former OpenAI researcher and RLHF co-inventor Diogo Almeida, launched Jev, an LLM that returns a defined decision plus its probability in a concise response instead of generating verbose…
Databricks reported a 60% increase in overall coding spend after rolling out GPT-6 Astra to roughly 3,500 engineers, even as benchmarks claim the model is cheaper than Sol on a per-token basis. OpenAI released a formal f…