The Allen Institute for AI (Ai2) released Olmo-core 3 on October 1st, a redesigned open training stack for mixture-of-experts models that Ai2 says reached 52,000 tokens per second per GPU on a 47-bill…
AppZen introduced ZenLM Plus on October 1st, a family of finance-specific language models that the company says outperformed seven frontier models on five of six expense-audit control comparisons, sco…
Modal made its multi-node GPU clusters generally available on October 1st, letting developers request coordinated GPU capacity with the Python decorator @modal.clustered(size=4, rdma=True) instead of …
Perplexity released a preview embedding model, pplx-embed-v2-context-9b-preview, on October 1 that encodes document chunks with the rest of the document in view, and the company says it leads on two r…
Context.dev launched Context Highlights on September 28th, a query-based web passage extractor that the company says used 97% fewer tokens than Codex in its own benchmark, reporting 76.0% groundedness…
Anthropic published an October 1st post describing BootLoops, an open-source toolkit built by Harvard physicist Matthew Schwartz that uses Claude to perform exact calculations across quantitative scie…
Anthropic announced on October 1st that Claude Code now supports developer-written "mods" that can rewrite prompts before they reach the model, block or retry tool calls, approve or deny permission re…
Tavus reported that its Griffin-Lite model was mistaken for a real person by 26 of 54 participants (48%) in one-minute live video calls, a study the company calls the first pass of a real-time video T…
Microsoft released three MAI voice models on October 1st: MAI-Transcribe-2-Streaming, MAI-Voice-2.1 and MAI-Voice-2.1-Flash, announced in a Microsoft AI post on X. The streaming transcription model ac…
Sidero Labs announced Talos Linux for AI and Talos Cupar on October 1st, pairing an open-source immutable operating system for GPU workloads with a commercial console for deploying and managing privat…
David AI's DAI-ASR-I18N benchmark, built by CEO Tomer Cohen and CTO Ben Wiley, compared 14 speech-recognition systems on unscripted conversation across 21 languages and found error rates varied by an …
HQ co-founder Jacob Posel said on September 30 that the company's shared AI context layer has officially launched and that nearly 1,000 businesses already run on HQ, a customer figure the company repo…
Carter Church published a proposed solution on September 18 to a 217-year-old cipher letter addressed to Napoleon's marshal Auguste de Marmont, saying GPT-6 Astra helped transcribe and decode the ciph…
Nick Khami launched Call4me on September 30th, an MCP-connected phone-calling service that lets AI agents including Claude Code and Codex call businesses to book appointments, change flights and cance…
LiteLLM launched Lens on September 30th, a tracing and analysis product that uses AI agents to find recurring problems across agent runs, according to co-founder and CTO Ishaan Jaffer. Lens stores tra…
Monid added a TinyFish-powered workflow that lets AI agents reuse authenticated browser sessions across runs, with LinkedIn, Amazon, Workday, Facebook and Instagram listed as examples, co-founder Shen…
CometAPI announced on September 30 that a single API key now provides access to more than 500 AI models, with CEO Lee Sonic saying the gateway is meant to spare developers from juggling separate regis…
Snorkel AI research fellow Gabriel Orlanski submitted a paper on September 29 introducing LibraryDesignBench, a benchmark that scores agent-generated libraries by how well downstream agents use them a…
Cohere released Embed 5 on September 30th, a pair of embedding models for enterprise search and retrieval that share an embedding space so customers can index with the higher-quality Pro model and que…
Hacktron co-founder and CTO Mohan Pedhapati reported that GPT-5.6 Sol Ultra built a working Chrome/V8 exploit chain in a controlled benchmark published July 14th, 2026, at a reported model-compute cos…