Gemma 4

From the start

Monday, September 21

1. NewsGoogle for Developers · YouTube11:00 PM UTCWatch
   Understand the Gemma 4 model family

   Learn about the Gemma 4 family of multimodal open source language models from Google. Google DeepMind’s Maarten Grootendorst walks through all five model sizes across four architectures, from the efficient E2B and E4B mo…

Tuesday, September 22

2. NewsGoogle for Developers · YouTube11:00 PM UTCWatch
   Gemma 4 12B: The encoder-free model explained

   Learn what makes Gemma 4 12B, Google's encoder-free open source language model, unique in the Gemma 4 family. Google DeepMind’s Maarten Grootendorst explains how it processes audio and image inputs without encoders. Subs…

Friday, September 25

3. NewsGoogle for Developers · YouTube4:00 PM UTCWatch
   Per-Layer Embeddings (PLE) in Gemma 4 explained

   Learn how per-layer embeddings, or PLE, work in Gemma 4 E2B and E4B, Google's open source language model. Google DeepMind’s Maarten Grootendorst breaks down how PLE gives each token a layer-specific embedding for more di…

Thursday, October 8

4. NewsGoogle for Developers · YouTube11:00 PM UTCWatch
   Building with Gemma 4

   Gemma 4 is the most capable model family you can run on your hardware. They complement our Gemini models, giving our developer community the industry's most powerful combination of both open and proprietary tools. Watch…

Monday, October 26

5. ReleaseOllama12:07 AM UTCReleases
   Ollama 0.40.12: mlx, preserve small-image detail for compact Gemma 4 vision towers by @dhiltgen in #19247

   What's Changed mlx: preserve small-image detail for compact Gemma 4 vision towers by @dhiltgen in #19247 MLX, MLX-C: version bump - prompt speed up for gemma/flint/laguna by @dhiltgen in #19291 mlx: add DFlash2 support b…

Monday, November 2

6. AnalysisDigitalOcean Tutorials5:51 PM UTCTechnical Guides
   DigitalOcean Serverless Inference latency: p50, p95 and p99 TTFT for GLM-5.3, DeepSeek V3.2, Gemma 4 and Llama 4 Maverick (3,456 requests, 48 hours)

   p50, p95 and p99 TTFT and tokens per second for GLM-5.3, DeepSeek V3.2, Gemma 4 and Llama 4 Maverick on DigitalOcean Serverless Inference, over 48 hours.

The story so far.