OpenAI opens Ultrafast mode for GPT-6.1 Sol to all API users
The speed tier promises up to 8x faster output than standard mode and costs 6x the standard rate, or $12 in and $60 out per million tokens.
Why it matters Developers can now pay for speed on the cheaper Sol model, not only on the flagship Astra, which matters for agents that make many quick tool calls.
Google Cloud’s Gemini agent takes goals, not step-by-step prompts
Announced at Gemini at Work, the agent plans multi-step jobs across Workspace, Slack and Microsoft 365 and picks a Gemini or Claude model for each step.
Why it matters Google is selling one work agent that also routes to a rival’s models, so the contest shifts from the best model to the best workplace layer.
State of AI 2026: Claude now leads a quarter of Anthropic’s R&D
Air Street’s ninth annual report says Claude led 26% of Anthropic’s measured model R&D work in August, up from under 1% in February, with people still setting and supervising the tasks.
Why it matters The yearly report is a quick checkpoint on how fast AI is speeding up its own development, and where public pushback is starting to bite.
Audit finds two-thirds of Xiaomi’s MiMo coding tasks leak the answer
Vals AI checked 2,698 coding tasks Xiaomi released from its MiMo v2.6 training and found the reference fix still sat in leftover Git objects in 1,795 of them.
Why it matters Training tasks that leak answers can teach coding agents to cheat, so open training environments need audits as much as models do.
Researchers say Lean checks can miss errors in AI-translated proofs
A new arXiv paper argues that when AI translates a written proof into Lean, the checked code can prove a different statement than the original.
Why it matters A machine check only builds trust if the formal statement says what the paper says, so AI math results still need people to review the translation.
Utah lets an AI app write first acne prescriptions
Nolla Health’s Nolla Derm app uses a face scan and questionnaire to pick from pre-approved topical acne treatments for Utah adults, for $4.99 a month.
Why it matters This is the first U.S. test of AI issuing a new prescription, not just a refill, so its safety record will shape how other states treat AI in routine care.
Isomorphic Labs in early funding talks at $40B or more
Bloomberg reports the DeepMind drug-discovery spinout seeks new funds at a valuation of at least $40 billion, possibly up to $50 billion, five months after a $2.1 billion round.
Why it matters Investors are paying for the promise of AI drug design, so the value now rests on whether its drug candidates succeed in clinical trials.
OpenAI and Liquid AI ship models that answer, not write
OpenAI’s Decisions API, now in public beta, uses GPT-6 Luna to return yes/no probabilities, picks and scores for $0.10 per million input tokens, with no output-token charges.
Why it matters Many app jobs, like spam checks, routing and part inspections, need a fast answer with a confidence level rather than a paragraph, so these models can cut cost and delay for agents and edge devices.
On a visual common-sense test, people score 93%, top AI 54%
Scale Labs and Elorian’s Humanity’s Sixth Sense has 522 open-ended tasks across 288 images and 234 video clips, such as whether a car fits a gap or who holds authority in a room.
Why it matters Robots and self-driving systems need this everyday intuition, so the gap shows where vision models still fall short for work in homes, streets and workplaces.
Crew-12 splashes down off Los Angeles after 237 days
Dragon landed in the Pacific at 11:34 a.m. EDT with Jessica Meir, Jack Hathaway, ESA’s Sophie Adenot and Roscosmos’ Andrey Fedyaev, after more than 3,792 orbits and four spacewalks.
Why it matters A clean return closes the rotation on schedule and keeps Dragon the steady crew ride to the station.
Blue Origin picks Texas for a $555M satellite factory campus
Constellation Park in Hutto will cover 1.3 million square feet and build hardware for TeraWave, Blue Origin’s planned satellite network for enterprise, data center and government users.
Why it matters A dedicated factory is a sign Blue Origin plans to build satellites at volume to compete with Starlink.
The archive holds crew photos, audio, tablet notes and a preliminary science report, including five faint impact flashes the crew saw from 46,000 miles away.
Why it matters Astronaut observations tied to exact images give scientists context that orbiters cannot, and the lessons shape how future Artemis crews will do science on the Moon.
First recovered Starship, Ship 40, arrives back at Starbase
Ship 40 flew Flight 13 on July 24, splashed down intact in the Indian Ocean, and reached Starbase on Oct. 7 aboard a heavy-lift ship after 75 days at sea.
Why it matters A flown upper stage on the ground gives engineers real wear data on the heat shield and structure, which feeds SpaceX’s goal of reusing Starships.
Mecka raises $60M to record everyday human work for robots
Sequoia led the Series B, with Nvidia, Microsoft’s M12, Qualcomm Ventures and Samsung joining.
Why it matters Robots lack internet-scale data on touch, force and motion, so firms that capture real human demonstrations may set how fast humanoids learn new jobs.
U.S. Department of War offers Wolfspeed a 30-year, $1.5B loan
The conditional offer from the Office of Strategic Capital would first refinance $600M of Wolfspeed debt, then fund U.S. silicon carbide and gallium nitride production plus radiation-hardened parts.
Why it matters These power and radio chips sit inside EVs, radar and data centers, so Washington is paying to keep a domestic source alive.
Broadcom seeks over $50B in private credit for OpenAI’s custom chip
The Wall Street Journal reports Broadcom is in early talks with lenders including Apollo and Blackstone to fund OpenAI’s “Nexus” chip program, with a close targeted by year-end.
Why it matters Private debt, not only cash, now pays for AI compute, which speeds the buildout but adds risk if AI revenue slows.
DARPA moves IBM, IonQ, Atom and Diraq to final quantum test stage
In Stage C of its Quantum Benchmarking Initiative, DARPA’s own team will test whether each company’s utility-scale design can be built and work as claimed.
Why it matters Independent government testing gives a rare outside check on quantum roadmaps that are otherwise hard to verify.