On October 9, Microsoft released Decision-1, a scoring model for structured decisions. It writes no text — it picks the best option from a given set and returns probability scores. Built on Alibaba's Qwen3.5-9B, input costs $0.042 per million tokens, output is free. The decision-model track that startups heated up last month now has Microsoft, OpenAI, and Cloudflare all in.
StepFun shipped its flagship Step 5 Preview — 600B total parameters, 27B active, 1M-token context, aimed at coding, finance, and long-horizon agent tasks. The API is live on OpenRouter; open weights expected October 15.
On October 6, Anthropic released the public beta of Claude for Google Workspace, adding a Claude sidebar to Docs, Sheets, and Slides that all paid subscribers can install. This first-party integration puts Claude inside Gemini's home turf.
On October 7, Musk announced on X that Grok Bot will no longer rely solely on in-house models. Complex reasoning goes to Claude Opus 5.5, image generation to Midjourney, music to Suno. Grok Bot has evolved from an in-house tech showcase into a routing layer that picks backend models per task.
At its October 8 event, Google Cloud unveiled a unified work agent: give it a goal and it plans, picks its own tools, and gets the job done across Gmail, Slack, Salesforce, and more. You can even create coworker agents with their own email addresses — close your laptop and the task keeps running in the cloud.
Manus parent Butterfly Effect announced a funding round of over $500 million, co-led by Boyu Capital and IDG Capital. It's the first round since the Meta acquisition was called off and the company resumed independent operations.
Microsoft will hold an event in San Francisco at 1 AM Beijing time on October 8, with CEO Satya Nadella and NVIDIA CEO Jensen Huang sharing the stage to unveil the Surface Laptop Ultra powered by the RTX Spark chip — up to 128GB of unified memory, capable of running 120B-parameter models locally. Microsoft wants to move AI agents from the cloud back onto local devices.
Nvidia-backed Reflection AI released its first open-weight model Beam on October 5: 501B total parameters with only 23B active per token, 1M-token context. The company claims reasoning performance close to Zhipu's GLM-5.2 at one-quarter to one-third the inference compute, with weights coming later in October under Apache 2.0.
Synopsys signed a multi-year deal with OpenAI to co-develop GPT-Synopsys, a specialized model that directly operates EDA tools. Synopsys stock jumped 12.78% on the news, and the company raised FY2027 revenue guidance to $11.0–11.2 billion.
On October 2, Meta open-sourced Muse Gadgets: ESP32 firmware and a Linux SDK under Apache 2.0, letting developers hook the Muse agent up to their own hardware. The companion reference device, Muse Home Link, got a 5,000-unit production run — free for US Muse subscribers.
On October 1, Cloudflare released the Clef and Clef-flash decision models: they generate no text at all, only outputting probabilities for three fixed question types. The 27B accuracy model and 9B latency model are open-weight under Apache 2.0, with an API compatible with Typesafe's Jev.
On September 22, Alibaba unveiled its full-stack AI strategy at the Apsara Conference in Hangzhou: Qwen 4 is training, future versions planned at 5–10 trillion parameters; its in-house Zhenwu V900 AI chip is 3x the previous generation, mass production in Q1 2027; Alibaba Cloud targets over 20GW of global data center capacity by 2032.
OpenAI unveiled dots at DevDay 2026: persistent AI agents. Each dot runs on GPT-6 Astra with its own cloud computer and browser, continuing tasks while you're away. ChatGPT Pro and Business Premium subscribers get first access.
On September 29 local time, OpenAI's 2026 developer conference brought 20+ updates: the always-on agent Dots works independently and connects 4,000+ apps, new model GPT-6.1 Sol launched, and team workspace ChatGPT Space debuted. Five minutes to the three highlights.