Monday 20 July 2026

The Brief – 21/07/2026

The top things worth knowing about in AI today.

  1. Moonshot pauses Kimi K3 signups

    Chinese lab Moonshot stopped taking new Kimi K3 subscriptions on 19 July after demand pushed its GPUs near capacity within 48 hours. K3 is a 2.8-trillion-parameter open model with a one-million-token context window; existing users keep access and new spots reopen in batches. Moonshot will split memberships into general and dedicated coding plans and release the full weights as open source by 27 July. The bottleneck for the fastest-moving Chinese labs is now compute, not model quality.

    Read more
  2. EU forces Google to open Android to rivals

    The European Commission ordered Google on 16 July to give rival AI assistants equal access to 11 Android features, from voice activation to acting inside apps, that had been reserved for Gemini. Google must also share anonymised search query, click and ranking data with competing search engines. The Android changes must ship in Android 18 by August 2027, with data sharing phasing in through 2026. For anyone building or choosing an AI assistant, it opens a real path to competing with the default position Gemini holds on phones.

    Read more
  3. Databricks raises at $188 billion valuation

    Databricks signed a term sheet for a roughly $3 billion round at a $188 billion valuation, led by existing investor Coatue and due to close this summer. That is up from $134 billion only months earlier. The capital funds its AI products, including Unity AI Gateway, Genie and Lakebase, as companies consolidate their data and AI tooling onto fewer platforms. It signals that investor appetite for the infrastructure layer of AI has not cooled.

    Read more
  4. AI labs chase enterprise deployment

    The Anthropic-backed venture Ode, funded by Blackstone, Goldman Sachs and others at $1.5 billion, has bought engineering firm Fractional AI and now embeds specialists inside client operations to make AI actually work. It joins the OpenAI Deployment Company, the $2.5 billion Microsoft Frontier unit with 6,000 experts, and the Meta Enterprise Solutions team. The shared bet is that the next big AI business is implementation, not the models. It reflects a real gap: 71% of executives at billion-dollar companies say readiness, not technology, is what holds their AI back.

    Read more
  5. Australia sets up national AI office

    The federal government announced an Office of Artificial Intelligence inside the Department of the Prime Minister and Cabinet on 15 July, alongside plans for a national AI framework it says would be the first legislated by any government. Writers, artists and journalists will keep ownership of their work, so companies cannot train AI on Australian creative material without consent from the creator. National Cabinet considers the plan in August, with standards expected to be legislated early next year. Businesses using AI should expect clearer, and enforceable, rules on data and content.

    Read more
  6. 1Password lets Claude sign in without passwords

    1Password launched an integration on 16 July that lets Claude sign in to websites without the model ever seeing the password, login item or one-time code. Credentials stay in the encrypted vault, and each login needs the user biometric approval; when an agent takes over the browser, the extension locks down to only the logins approved for that task. It removes a practical blocker to letting agents act on real accounts. Available now for Mac users on individual, family and business plans.

    Read more
  7. Google delays Gemini 3.5 Pro again

    Google has again pushed back Gemini 3.5 Pro, its answer to GPT-5.6 and Claude, after the rebuilt model fell short of internal targets on hallucination rates and reliability, Bloomberg reported. The model, a full rebuild with a reported two-million-token context window, is being tested by a handful of enterprise customers but has no public release date or model card. Prediction markets now expect late July or early August. The delay is a reminder that even the largest labs are hitting reliability ceilings, not just chasing bigger context windows.

    Read more
The Weekly AI Brief

Practical AI, distilled.

A short read every week — the few things worth your time, and nothing that isn’t.