Meta releases Llama 4 as open source with MoE architecture
Meta is launching its new AI model series, Llama 4, as open source. For the first time, the models introduce a Mixture of Experts (MoE) architecture to the Llama family, along with a context window of up to 10 million tokens.
What happened?
Meta has launched its new Llama 4 model series as open source. For the first time in the Llama series, Meta is utilizing a Mixture of Experts (MoE) architecture, meaning only a subset of parameters is activated per token during inference. The series includes the models Scout (109B total parameters, 17B active), Maverick (400B total parameters, 17B active), and Behemoth, which is currently still in training. The Scout model supports a context window of up to 10 million tokens via iRoPE architecture.
Key facts
| Scout totala parametrar | 109B (17B aktiva) |
|---|---|
| Maverick totala parametrar | 400B (17B aktiva) |
| Scout kontextfönster | 10M tokens |
| Maverick ELO på LMArena | 1417 |
Why it matters
The shift to an MoE architecture allows for maintaining high knowledge capacity within the model while significantly reducing computational requirements and latency during inference compared to dense models. By activating only 17 billion parameters per token, models with hundreds of billions of total parameters can be run more efficiently on existing hardware. Test results from LMArena also indicate that Maverick achieves high scores among open models.
Who is affected?
Developers, AI researchers, and companies building proprietary AI services are directly affected, as they gain access to open MoE models with lower computational costs for inference. Organizations managing large document repositories or extensive codebases can leverage the expanded context window. End users will be affected indirectly as applications and services integrate these new models.
Impact on the EU
Llama 4 is being released as open source globally, but companies within the EU must ensure that their implementations comply with the regulations set forth in the EU AI Act and GDPR regarding data protection and transparency when handling personal data.
What else you should know
The report on Llama 4 originates from Chinese media reports and refers to test results on the LMArena platform. It is important to note that date information in the source material appears inaccurate and that independent verification of the models' actual performance in production environments is ongoing.
Quick answers about this story
Vad har hänt?
När hände det?
Varför spelar det roll?
Påverkar det EU?
The link opens in a new window and leads to the publisher's own site.
Källan har spårats automatiskt från utgivaren via Aheadlines signalkedja.
Get similar news straight to your inbox
The reader's room
Send in a question or an addition. The newsroom reads everything before it's published and replies when relevant. No AI-generated text – just people.
Sign in to submit a comment or question.
Read the article through your role
- Decide whether this affects strategy over 6–12 months or is just noise.
- Discuss with leadership: do we own the right question or does ownership need to move?
- Ask: what risk are we taking by NOT acting on this this quarter?
Generated angle — not editorial analysis of "Meta releases Llama 4 as open source with MoE architecture"