Open weights

A model whose trained parameters are published, so anyone can download and run it on their own hardware.

Open weights are not the same as open source. The numbers are published, and often a license setting out what may be done with them, but the training data and the code that produced them usually are not, so the result cannot be independently reproduced. What the practice does change is control: a downloaded model runs without an internet connection, cannot be withdrawn by the company that made it, and cannot be modified after the fact.

Where it mattered in the record

  1. Reflection AI launches Beam

    On 5 October 2026 Reflection AI introduced Beam, its 501B-parameter open-weight model with 23B active parameters and 1M tokens of effective context. Beam stays in final red-teaming; weights will follow this month.

  2. Aleph Alpha releases open-weight model Kolibri

    On 3 October 2026 Aleph Alpha released Kolibri, an English-German mixture-of-experts Transformer with 78.1B total parameters and 3.46B active, under Apache 2.0 on Hugging Face. Context reaches 1M tokens, and Merlin-Arthur withholds answers the context does not support.

  3. Xiaomi releases open-weight MiMo-V2.6

    On 22 September 2026 Xiaomi released MiMo-V2.6 as open weights, continuing its push into frontier-class open models.

  4. Mistral raises €3 billion to fund sovereign AI ambitions

    On 8 September 2026 Mistral AI raised €3 billion in a round led by Samsung, valuing the French company at more than €21 billion as it positioned open-weight models as Europe’s sovereign AI answer.

  5. Meta releases Muse Glimmer under an Apache 2.0 license

    Meta released Muse Glimmer on 10 August 2026, a 30-billion-parameter multimodal model distilled from its closed Muse Spark model and published on Hugging Face under Apache 2.0. Compressed to about 4-bit precision, it runs on a single consumer GPU or Mac under 20GB.

  6. Alibaba announces Qwen3.8-Max, its largest model to date

    Alibaba announced Qwen3.8-Max on 3 August 2026 with 2.4 trillion parameters, about 95 billion active at a time through a sparse mixture of experts, a context window up to 1 million tokens, and native text, image and video input. Weights were scheduled to follow a week later.

  7. Anthropic sets out its position on open-weight models

    Anthropic published its position on open-weight AI models on 27 July 2026. Chief executive Dario Amodei said open-weight models without dangerous capabilities are a public good; the company backed chip export limits and mandatory safety testing rather than a ban.

  8. DeepSeek publishes its V4 model series under an MIT license

    DeepSeek published a preview of its V4 series on 26 April 2026, comprising V4-Pro at 1.6 trillion total parameters with 49 billion active and V4-Flash at 284 billion with 13 billion active, released under the MIT license with checkpoints on Hugging Face.

Every term in the glossary