Skip to content
AI.info

Model Tracker

Gemma 4

Gemma 4 is an AI model made by Google DeepMind. Google DeepMind released it on 2 April 2026. It is generally available. Its weights are open.

Google DeepMind

Gemma 4 is an AI model made by Google DeepMind. Google DeepMind released it on 2 April 2026. It is generally available. Its weights are open. Its licence is Apache 2.0.

Released
2 April 2026 — blog.google, read 16 October 2026
Status
Generally available — deepmind.google, read 16 October 2026
Takes
Text, Image, Audio, Video — ai.google.dev, read 16 October 2026
Returns
Not confirmed: two readings disagree: text, code and text
Open weights
Yes — ai.google.dev, read 16 October 2026
Licence
Apache 2.0 — ai.google.dev, read 16 October 2026
Context window
Not confirmed: The context length varies by variant, so the family has no single standard context-window figure. Source: https://ai.google.dev/gemma/docs/core/model_card_4 ; quote: " * Long Context – Context windows of up to 128K tokens (E2B/E4B) and 256K tokens (12B/26B A4B/31B)."
Longest answer
Not confirmed: The maker's Gemma 4 model card prints context lengths but does not state a standard maximum output-token limit. Source: https://ai.google.dev/gemma/docs/core/model_card_4 ; quote: "Context Length | 128K tokens | 128K tokens | 256K tokens | 256K tokens"
Knowledge cutoff
Not confirmed: The maker states a cutoff for the pre-training dataset, not a knowledge-cutoff date for the model; treating them as the same would be an inference. Source: https://ai.google.dev/gemma/docs/core/model_card_4 ; quote: "Our pre-training dataset is a large-scale, diverse collection of data encompassing a wide range of domains and modalities, which includes web documents, code, images, audio, with a cutoff date of January 2025."
API price, input
Not confirmed: The live API pricing page lists input as free of charge in the free tier and says a paid-tier price is not available; it does not print a standard paid input price to record. Source: https://ai.google.dev/gemini-api/docs/pricing ; quote: "Input price | Free of charge | Not available"
API price, output
Not confirmed: The live API pricing page lists output as free of charge in the free tier and says a paid-tier price is not available; it does not print a standard paid output price to record. Source: https://ai.google.dev/gemini-api/docs/pricing ; quote: "Output price | Free of charge | Not available"
Where
Not confirmed: two readings disagree: api, weights, cloud and api, weights
Parameters
Not confirmed: Gemma 4 is a family with multiple parameter sizes, not one model parameter count. Source: https://ai.google.dev/gemma/docs/core/model_card_4 ; quote: "The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B."
Model card
https://ai.google.dev/gemma/docs/core/model_card_4 — ai.google.dev, read 16 October 2026

Questions

When was Gemma 4 released?

Google DeepMind released Gemma 4 on 2 April 2026.

Is Gemma 4 still available?

Yes: it is generally available, as Google DeepMind states it.

Are the weights of Gemma 4 open?

Yes: Google DeepMind publishes the weights of Gemma 4 for download.

What is the licence of Gemma 4?

The licence of Gemma 4 is Apache 2.0, as Google DeepMind names it.

More from Google DeepMind

Sources

  1. Gemma 4 model card (read 16 October 2026)
  2. Gemma 4: Byte for byte, the most capable open models (read 16 October 2026)
  3. Model cards (read 16 October 2026)
  4. Gemini Developer API pricing (read 16 October 2026)
  5. Release notes (read 16 October 2026)
  6. Deploy Gemma with Google Cloud (read 16 October 2026)
  7. Gemma 4 (read 16 October 2026)

All of the Model Tracker