Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Anthropic's Frontier Red Team open-sourced FlashProve on November 2, 2026, a research prototype that generates zkSNARK proofs binding an output token sequence to a specific committed set of model weights without revealin…
MV2 of Munim, Inc. coded 44 federal protective orders governing AI use in discovery and tested six pre-registered predictions about whether they permit running open-weight models on a firm-controlled server. Three predic…
Tencent launched Hy4 on Wednesday, a large language model available through hosted APIs that adds visual understanding and computer-use capabilities, with model weights planned for release in the coming days. The release…
Kong released Kong AI Observability, a product that traces a full AI task across models, tools, agents and policies within the connectivity path rather than assembling telemetry from separate systems, according to a post…
An independent open-source project called One Code recreates Claude Code's coding workflow on top of pi while letting users plug in any model or provider, released free under the MIT license as the npm package @one-ai/on…
Google Cloud updated its Agentic Data Cloud platform with a borderless Lakehouse built on Apache Iceberg that caches data from Amazon Web Services and Microsoft Azure over private links without egress fees, plus an Alloy…
Google announced new capabilities for its Agentic Data Cloud, including Gemini working directly on the borderless Lakehouse so business users can query records across clouds in natural language, and cross-cloud interconn…
Two studies dated 26 Oct found that deleting old tool output from coding-agent runs can raise costs despite shortening prompts: keeping only the last five tool outputs made prompts 33.3% shorter but increased uncached pr…
A Dev.to tutorial by Rashvan Dhappukutty walks developers through building a local-first AI assistant with Ollama, covering pip installation of the ollama library, downloading a pre-trained model locally, and making text…
A developer at Stepthrough built a support chatbot on Dify that answers member questions from a single Markdown help file, and documented the failure modes found in testing. Splitting the file on level-2 headings instead…
A UC Berkeley and Microsoft Research team released TokenEconomizer, a plugin for Claude Code, Codex, the GitHub Copilot CLI and Pi that cut mean coding-agent cost by 18.4–30.9% across five agent–model combinations on 277…
Duolingo's DevX AI team built AI STR Fix, an agent using the Claude Agent SDK with Sonnet handling most tasks and Opus taking harder ones, to automate the investigation work behind Shake-to-Report bug tickets that automa…
Three research groups publishing between 21 and 26 October found that elaborate agent memory systems underperform plain retrieval-augmented generation over raw chat history, with He et al. at UIUC and Tsinghua naming the…
Towards Data Science published a post titled "Let Decision Models Guide Your Voice Assistant" describing an architecture in which GPT-Live handles speech while a separate decision model tracks which items are confirmed a…
Tencent released Hunyuan Hy4 on November 4 with vision and computer-use capabilities, upgrading the August 28 text-only 770-billion-parameter mixture-of-experts preview into a model that can read charts, documents and mu…
A developer built Slothpress, a client-side quote-card app that runs a quantized 25 MB MobileBERT checkpoint (Xenova/mobilebert-uncased-mnli) via transformers.js and ONNX Runtime Web to rank eight typography presets agai…
Anthropic, OpenAI and Google all price batch LLM inference at 50% of standard rates as of November 2026, with results taking up to 24 hours, according to a comparison of the three providers' published pricing. For a work…
Trida AI published a blog post arguing that AI agent systems should attach claim-level "receipts" to answers rather than relying on run traces alone. The post proposes a four-field receipt format — exact source, passage …
Rezi.ai moved its entire 515-page website off Webflow in one weekend using Anthropic's Claude Opus 5.5, after Webflow quoted $2,500 a year to add a single language on top of a roughly $30,000-per-year enterprise contract…
Anthropic's Claude API prompt caching lets developers reuse an unchanged prompt prefix at a cache-read rate generally reported at 0.1x of the base input price, with a 5-minute cache write at 1.25x and a 1-hour write at 2…