Thursday, 8 October 2026 · Daily Briefing

Pulse AI Briefing

OpenAI’s GPT-6 with Intelligent UI now reaches all of ChatGPT’s 1.2 billion weekly users, turning answers into charts, buttons and mini-apps — as Anthropic cuts the price of a capable small model by around 75%, and a record $80bn quarter at Samsung shows where the build-out money is landing.

The signal

OpenAI puts a billion users on GPT-6 — and an interface that builds itself

OpenAI began rolling GPT-6 into ChatGPT on Wednesday for Plus, Pro, Business and Enterprise, and from today it reaches the Free and Go tiers — the more than 1.2 billion people who use the chatbot each week. The model arrives with Intelligent UI, which composes answers from text, visuals and interactive elements: side-by-side comparisons, editable diagrams, charts, calculators and small tools built for the question at hand, with the format chosen automatically and a plain-text answer still available when that works best. Paid tiers run GPT-6 Sol; free tiers run GPT-6 Luna. OpenAI frames the change as a reversal of who adapts to whom — “instead of people adapting to software, software will adapt to people”. It lands in the same week that Anthropic cut the cost of running its small model by around 75%. Source: OpenAI · TechCrunch

Seven stories

Today’s Briefing

Models · 7 October

Anthropic cuts its cheapest model to a fraction of last year’s price

Anthropic has launched Claude Haiku 5.5, which it calls its fastest and most capable small model yet — and roughly 75% cheaper to run than Haiku 4.5 on average, or 90% cheaper on prompts up to 100,000 tokens, which it says make up about nine in ten requests. Input costs $0.10 per million tokens and output $0.50, matching OpenAI’s budget GPT-6 Luna. Anthropic also halved the price of Sonnet 5.5’s cache reads, trimming the cost of most agentic work by around a fifth, and added monthly API credits for Max and Team subscribers. The pitch is a barbell: large models for hard work, a fast cheap one to run beneath them.

Source: AnthropicYahoo Finance

Hardware · 7 October

Microsoft ships Nvidia-powered AI PCs, and a $6,000 desktop for local models

At its first major Windows and Surface event in more than two years, Microsoft revealed prices and specs for the Surface Laptop Ultra, an AI PC built on Nvidia’s RTX Spark chip, starting at $2,600 and rising to $5,900. Its companion, the Surface RTX Spark Dev Box workstation, starts at about $6,000 and ships with VS Code, GitHub Copilot CLI, WSL and PowerShell 7 — aimed at developers who want to run 120-billion-parameter models on the device rather than in the cloud. A revamped Windows 11 adds “Execution Containers” to sandbox AI agents, available to all users. Nvidia’s Jensen Huang shared the stage with Satya Nadella, who said a model alone “doesn’t do much” without orchestration, memory and an action space.

Source: TechCrunchThe Information

Economy · 8 October

Samsung posts an $80bn quarter as the AI memory boom runs hot

Samsung Electronics expects a third-quarter operating profit of 107.4 trillion won (about $80.2bn), it said on Thursday — a roughly ninefold rise on a year earlier and the first time any technology company has topped 100 trillion won in a quarter. Revenue is put at around 195 trillion won. The driver is high-bandwidth memory and other chips for AI infrastructure, where demand continues to outrun supply and has lifted prices across the market; SK Hynix and Micron have ridden the same wave. It is the clearest financial signal yet that the binding constraint on AI is not appetite but hardware.

Source: ReutersCNBC

Infrastructure · 7 October

Broadcom lines up $50bn for OpenAI’s custom chips

Broadcom is working to arrange more than $50bn in financing for the custom AI chip it is developing with OpenAI, with Apollo and Blackstone among the lenders approached and a target of closing before year-end, the Wall Street Journal reported. Oracle is separately in talks with Apollo and Goldman Sachs to fund a large chip purchase through an off-balance-sheet entity, and SpaceX has discussed about $40bn — roughly $10bn in bank loans and $30bn in investment-grade bonds — for Nvidia chips. Together the three mark a turn from public bond markets and cloud leasing towards private credit to pay for the build-out.

Source: WSJ

Policy · 7 October

Finland halts Google’s data centres over missed reviews

Finland’s licensing authority, LVV, has ordered Google subsidiary Tuike Finland to suspend land-altering work at two planned data-centre sites in Muhos and Kajaani until environmental impact assessments are completed — a month after Google committed $15bn to AI infrastructure in the country. The stop order, due no later than 23 October, covers tree felling, excavation, quarrying and road building; roughly 330 hectares at Muhos and 200 at Kajaani had already been cleared. “We understand the concerns and have fallen short of our own high standards in this instance,” a Google spokesperson said. It is an early test of the backlash against the physical footprint of AI.

Source: CNBC

Funding · 7 October

Nous Research raises $90m for its open-source agent

Nous Research, the lab behind the MIT-licensed Hermes Agent, has confirmed a $90m Series B at a $1.5bn valuation, first reported by the Wall Street Journal. The capital funds a push into the enterprise with “Hermes for Businesses”, letting companies deploy customised agents for multi-step workflows while keeping their data private. It is a rare open-weights project raising at scale, and a counterpoint to the closed frontier labs — and a reminder that the agentic layer, not just the model, is now attracting capital.

Source: TechCrunchNous Research

Geopolitics · 7 October

China’s labs ship 16 models in a month, defying calls to slow

Chinese developers including DeepSeek, Alibaba and Xiaomi released 16 AI models in September alone, Nikkei Asia reports, even as Anthropic chief executive Dario Amodei urged the industry to pause frontier work over safety risks. Nikkei calculates the average interval between significant model releases has compressed from about 125 days between January 2023 and March 2026 to 44 days from April to September 2026, across nine leading US and Chinese labs. DeepSeek has shipped updates every month since July; Alibaba and Z.ai followed. A DeepSeek engineer told the paper he did not trust the American labs to keep advanced AI open and affordable.

Source: Nikkei Asia

Calendar

What to Watch

  • Today

    State of AI Report 2026 lands

    Nathan Benaich and Air Street Capital publish the ninth annual State of AI Report, their open-access survey of AI research, industry, geopolitics and safety, with launch meetups in San Francisco and beyond. It has become the sector’s de facto annual scorecard.

  • 13–15 Oct

    TechCrunch Disrupt 2026, San Francisco

    TechCrunch’s flagship startup conference returns, with its Real World AI stage turning to what happens when agents move from prototype to production — the operational counterweight to this cycle’s launches.

  • 22–23 Oct

    AGNTCon + MCPCon North America, San Jose

    The Agentic AI Foundation’s flagship gathering for the open agentic stack — the Model Context Protocol, the goose framework and the AGENTS.md spec — covering how teams run agents in production: reliability, permissions, security and observability.

  • 27 Oct

    Mistral’s Large 4 weights due

    Mistral says it will publish the open weights for its trillion-parameter Large 4 by the end of October, with reporters told the 27th. Until then, every performance claim about the model remains self-reported.

  • 9–12 Nov

    Web Summit 2026, Lisbon

    Europe’s largest tech gathering returns to Lisbon with 70,000-plus attendees, and an AI track that is now among its biggest. The venue where the year’s agentic and infrastructure themes meet their market.