Anthropic commits $100 million to train 10,000 Frontier Deployed Engineers
Anthropic announced Claude Frontier Academy, a $100 million program to train 10,000 engineers by the end of 2027, aimed at the shortage of people who can deploy AI inside large enterprises. It targets experienced engineers from firms such as Accenture, Bain, Capgemini, Deloitte, McKinsey and Morgan Stanley. The format follows a medical-residency model: a multi-day in-person bootcamp, then a 12-week Claude project led at the participant's own company with Anthropic engineer support, and a credential on completion. It is a talent and go-to-market move, not a model release.
Sources & depth
NVIDIA's DGX Spark gets a 64GB version at $4,999, shipping October 23
NVIDIA introduced a 64GB DGX Spark configuration, available from October 23, 2026 through six manufacturers starting at $4,999. It uses the same Grace Blackwell compute, unified memory and ConnectX-7 networking, and NVIDIA says it runs models up to 100 billion parameters entirely on-device. Two 64GB units can be clustered to 128GB with the Sync Cluster Assistant, for up to 1.7x the performance of a single unit. It lowers the entry price for desktop-class local inference.
Sources & depth
Ai2 open-sources AstaBrief 8B, a fast cited-report model
Ai2 released AstaBrief, an 8-billion-parameter open-weights model that turns a research question plus retrieved literature excerpts into a cited report. It reportedly generates a report in about 51 seconds, roughly 3.5x faster than proprietary alternatives, with competitive quality on scientific-synthesis benchmarks (SQABench-CS2, DeepScholarBench). Training data was filtered for citation density, and evaluation scores answer precision plus citation precision and recall. Speed and quality figures are Ai2's own.
Sources & depth
Ollama 0.35.1 adds Cloudflare's Clef decision models and Modelfile capability declarations
Ollama now serves Cloudflare's open-source Clef (27B) and Clef Flash (9B) decision models through a new /v1/systemone endpoint. Both are multimodal, so requests can include images alongside the text state. The release also lets Modelfile authors declare model capabilities explicitly, preserved through inheritance and export, raises web search to up to ten searches per response, and updates the llama.cpp and MLX engines.
Sources & depth
- github.comv0.35.12026-10-02 23:58 IST
OpenAI publishes a model-selection guide for the GPT-6 family
OpenAI posted a startup-focused guide on choosing between GPT-6 models, tuning reasoning effort, improving prompts and skills, coordinating tools and preparing workflows for production. The page itself could not be opened, so this rests on OpenAI's own summary line only — link only.
Sources & depth