Meta Ads · Status at source date: Newsletter archive
Google launches Gemma 4, its most capable open-source agentic model
Google has released Gemma 4, a family of fully open-source models under Apache 2.0 built on the same research as Gemini 3. The family spans four sizes: 2B (on-device), a 26B Mixture of Experts model (activating 3.8B parameters during inference), and larger models up to 31B dense: all with native multimodal reasoning across video, images, OCR, and charts.
What changed
All models support agentic workflows with built-in function calling, structured JSON output, and system instructions, and offer context windows of 128K to 256K tokens. The 26B MoE model is designed for workstation deployment; the 2B and 4B edge models run offline on phones and IoT devices. Available immediately on Google AI Studio, Vertex AI, Hugging Face, and Ollama.
Why it matters for advertisers
For you, this means that enterprise and agency teams building custom advertising automation or reporting tools have access to a capable, commercially free model with agentic capabilities: without API costs or data-sharing constraints.
Perspective from the original PMC newsletter.Sources & contributor credit
- Newsletter coverage · Paid Media Collective newsletter
Original newsletter text, contributor labels and media for this update.
- Source referenced in newsletterLinkedIn
Linked from the original newsletter. The source publication date has not been independently confirmed.
Original creator unverified
The names below are source credits. The original creator and first-reporting priority have not yet been independently verified.
Attribution evidence and limitations
Full attribution review pending.
A source link alone does not establish original authorship.
Original image and video creators have not yet been verified.
- Published on this site
This update reflects the dated source reporting. Availability may have changed. Further coverage of this same development will be added to this page.
Original newsletter text and archive evidence
Google launches Gemma 4, its most capable open-source agentic model
Google has released Gemma 4, a family of fully open-source models under Apache 2.0 built on the same research as Gemini 3. The family spans four sizes: 2B (on-device), a 26B Mixture of Experts model (activating 3.8B parameters during inference), and larger models up to 31B dense: all with native multimodal reasoning across video, images, OCR, and charts. All models support agentic workflows with built-in function calling, structured JSON output, and system instructions, and offer context windows of 128K to 256K tokens. The 26B MoE model is designed for workstation deployment; the 2B and 4B edge models run offline on phones and IoT devices. Available immediately on Google AI Studio, Vertex AI, Hugging Face, and Ollama.
For you, this means that enterprise and agency teams building custom advertising automation or reporting tools have access to a capable, commercially free model with agentic capabilities: without API costs or data-sharing constraints.
Source captured . No explicit first-contributor label was provided for this update.
