AI news

New AI models, products and features, written from each company's own announcement and linked to it. What happened, and why it matters for the tools you use.

18 items · 6 linked to tools we list

  1. Hugging Face ·

    Reflection's Beam model arrives on Hugging Face

    On October 16, 2026, Hugging Face said Beam, the first model from US lab Reflection, is now on its platform.

  2. Hugging Face ·

    IBM releases Granite Switch 4.1 models in 8B and 30B sizes

    On October 15, 2026, the ibm-granite organization unveiled Granite Switch 4.1 models at 8B and 30B scales, which can shift efficiently between…

  3. GitHub ·

    GitHub AI Scan Reaches General Availability for Pull Requests

    On October 15, 2026, GitHub said its AI Scan feature for pull requests had moved to general release.

    GitHub Copilot

  4. GitHub ·

    GitHub makes agentic autofix for code scanning generally available

    On October 15, 2026, GitHub announced broad availability of agentic autofix for code scanning alerts.

    GitHub Copilot

  5. Hugging Face ·

    Apertus 1.5 gains image and audio understanding

    An entry in the swiss-ai organisation's Hugging Face blog space, dated October 15, 2026, unveiled Apertus 1.5 as a refresh of the 8B and 70B Apertus 1.0 models.

  6. Hugging Face ·

    Hugging Face Hosts Meta's Open Muse Spark Model Checkpoints

    On October 15, 2026, Hugging Face revealed Muse Spark Open, an open-weight mixture-of-experts model built by Meta.

    Muse Spark Open

  7. Google DeepMind ·

    DeepMind unveils on-device Gemma scam detector with Singapore govt

    On October 14, 2026, Google DeepMind revealed a privacy-focused scam detector that runs locally on hardware.

  8. Hugging Face ·

    LiquidAI launches open-weight LFM2.5-24B-A2B reasoning model

    On October 13, 2026, LiquidAI unveiled LFM2.5-24B-A2B, a reasoning model with 24 billion parameters, of which 2 billion are active, designed for quick,…

    LFM2.5-24B-A2B

  9. Google ·

    Gemini 4 Argon Opens Worldwide to Developers, Enterprises, and Subscribers

    On October 13, 2026, Google's newest frontier model, Gemini 4 Argon, rolled out worldwide.

    Gemini 4 Argon

  10. Hugging Face ·

    Falcon-ASR, Arabic Speech Model with Emirati Focus, Debuts

    Falcon-ASR, a speech-to-text system with 1.6 billion parameters, was unveiled on October 7, 2026.

  11. Google ·

    Google launches Playground, a platform for building custom games

    On October 7, 2026, Google unveiled Playground, an experimental platform where people can make, play, and share custom games.

  12. Google DeepMind ·

    Google DeepMind unveils EmbeddingGemma 2 embedding model

    On October 6, 2026, Google DeepMind rolled out EmbeddingGemma 2, expanding past text so one shared embedding space covers code, pictures, clips, and sound.

  13. Google DeepMind ·

    Google DeepMind debuts Gemini 4 Argon for cyber defenders

    On September 30, 2026, Google DeepMind unveiled Gemini 4 Argon, its latest frontier model.

    Gemini 4 Argon

  14. Google DeepMind ·

    SynthID Bio brings watermarking to synthetic biology

    Introduced on September 30, 2026, SynthID Bio extends watermarking into synthetic biology.

  15. Hugging Face ·

    Hugging Face Hub gains dedicated RL Environments area

    On September 28, 2026, Hugging Face said its Hub now has a dedicated spot for RL Environments.

  16. Google DeepMind ·

    Gemini 3.8 Live with Live Avatar for Gemini Enterprise

    On September 24, 2026, Google DeepMind introduced Gemini 3.8 Live with Live Avatar, which brings near real-time visual presence to its native live dialogue…

  17. Google DeepMind ·

    Google Private AI Compute to gain private server-side memory

    On September 23, 2026, Google's Private AI Compute Team outlined plans to add private, server-side memory to its Private AI Compute platform, which already…

  18. Hugging Face ·

    Hugging Face brings efficient GGUF model running to transformers

    On September 22, 2026, Hugging Face announced that transformers can now run GGUF models efficiently, letting people use checkpoints sized to fit laptop memory.