Menu

Models and launches

New AI models and product launches, from each company's own announcement.

Updated 24 Oct, 7 PM ET

  1. Moonshot AI releases Kimi K3.1 Pro, a model aimed at coding and agent work (opens the original source in a new tab)

    Moonshot says K3.1 Pro can read up to 1 million tokens (word pieces) at once and nearly matches Claude Opus 5.5 on a coding test. It is out now in the Kimi apps and API, with the weights promised by 4 November.

    Moonshot AI · Original source (opens in a new tab)

  2. Block unveils Dessa, a voice model trained only on recordings of people talking (opens the original source in a new tab)

    Block says it trained Dessa from scratch on human conversation alone, with no text and no pretrained models, so it can hold a live spoken chat full of pauses and emotion. Block has opened a waitlist.

    Block · Original source (opens in a new tab)

  3. OpenAI makes its Decisions API generally available (opens the original source in a new tab)

    The API reads text, images or JSON and returns a fixed-form answer, such as the chance something is true, one pick from a list or a score, which OpenAI says comes about 10 times faster than its Responses API.

    OpenAI · Original source (opens in a new tab)

  4. Anthropic opens a preview of Claude Duet, an app where Claude works on its own (opens the original source in a new tab)

    Anthropic says the app brings you what needs your attention, keeps doing tasks while you are away and checks with you before it acts. It opens by waitlist, first for US Pro and Max subscribers.

    Anthropic · Original source (opens in a new tab)

  5. Inception releases Mercury 3, a fast diffusion language model (opens the original source in a new tab)

    Inception says Mercury 3, a diffusion-based language model, beats GPT-6 Luna and Claude Haiku 5.5 on a range of tests while running 4 to 8 times faster, at over 1,000 tokens per second.

    Inception · Original source (opens in a new tab)

  6. MiniMax releases M3.1 Flash, a faster and cheaper model for AI agents (opens the original source in a new tab)

    MiniMax says M3.1 Flash takes a big step up in agent tasks, coding and computer use, runs nearly three times faster online, and costs 70% less to use than the model before it.

    MiniMax · Original source (opens in a new tab)

  7. Mistral releases Voxtral 3, speech models plus a kit for building voice agents (opens the original source in a new tab)

    The new models transcribe more than 50 languages and speak with changing emotion and pace, and the kit lets voice agents take actions during live calls. They run in Mistral's API or on your own servers.

    Mistral AI · Original source (opens in a new tab)

  8. Gemini 4 Argon is now available to developers and in the Gemini app (opens the original source in a new tab)

    Google's model for long, hard tasks such as coding launches at an introductory price of $2 per million input tokens. Google AI Pro and Ultra subscribers can use it in the Gemini app.

    Google · Original source (opens in a new tab)

  9. SpaceXSI releases Grok Voice 3.0, its new voice model (opens the original source in a new tab)

    SpaceXSI calls Grok Voice 3.0 the world's most intelligent speech-to-speech model, one that hears speech and answers in speech. It is in the API today at the same price as the version before it.

    SpaceXSI · Original source (opens in a new tab)

  10. Anthropic releases Claude Haiku 5.5 (opens the original source in a new tab)

    Anthropic says its new small model is its fastest so far and costs around 75% less to run than Haiku 4.5. It is also the first Haiku model with an adjustable effort setting.

    Anthropic · Original source (opens in a new tab)