Google DeepMind, "Gemma 3 Technical Report" (2025).

web

ai.google.dev·ai.google.dev/gemma

Gemma 3 is Google DeepMind's open-weight model family; this technical report is relevant to debates about open-weight model release norms, safety evaluations at deployment, and the proliferation of capable frontier-adjacent models.

Metadata

Importance: 45/100tool pageprimary source

Summary

Google DeepMind's technical report on Gemma 3, a family of open-weight language models designed for broad accessibility and deployment. The report covers model architecture, training methodology, capabilities benchmarks, and safety evaluations. Gemma 3 represents Google's contribution to open-model development with integrated safety measures.

Key Points

•Introduces Gemma 3 as a series of open-weight models varying in size, intended for research and commercial deployment
•Includes safety evaluations and responsible AI guidelines as part of the model release documentation
•Benchmarks demonstrate competitive performance across reasoning, coding, and multimodal tasks
•Discusses model fine-tuning capabilities and alignment techniques applied during training
•Relevant to AI safety discussions around open-weight model release practices and associated risks/benefits

Cited by 1 page

Page	Type	Quality
Large Language Models	Capability	60.0

Cached Content Preview

HTTP 200Fetched Apr 9, 20266 KB

Gemma — Google DeepMind Skip to main content Slide 1 of 3 Gemma

 Our most capable open models

 Learn more Try in Google AI Studio Read docs Your browser does not support the video tag. Your browser does not support the video tag. Maximum compute and memory efficiency

 A new level of intelligence for mobile and IoT devices

 Learn more Your browser does not support the video tag. Your browser does not support the video tag. Unprecedented intelligence-per-parameter

 Frontier intelligence on personal computers

 Try in Google AI Studio Learn more Your browser does not support the video tag. Your browser does not support the video tag. Explore the latest Our most advanced open models help developers create AI applications that run wherever users need them — from cloud servers to laptops and even phones. 

 Models 
 Variants 
 Gemmaverse 
 Integrations 
 Slide 1 of 11 April 2026 Introducing Gemma 4

 Byte for byte, the most capable open models. Purpose-built for advanced reasoning and agentic workflows

 Read blog January 2026 Introducing MedGemma 1.5 4B

 High-dimensional medical imaging interpretation in a compute-efficient model from Health AI Developer Foundations

 Read blog January 2026 Introducing TranslateGemma

 A collection of translation models to help people communicate across 55 languages

 Read blog December 2025 Introducing Gemma Scope 2

 Helping the AI safety community deepen understanding of complex language model behavior

 Read blog December 2025 Introducing FunctionGemma

 Bringing bespoke function calling to the edge

 Read blog December 2025 Introducing T5Gemma 2

 The next generation of encoder-decoder models

 Read blog September 2025 Introducing VaultGemma

 The world's most capable differentially private LLM

 Read blog Your browser does not support the video tag. Your browser does not support the video tag. September 2025 Introducing EmbeddingGemma

 The best-in-class open model for on-device embeddings

 Read blog August 2025 Introducing Gemma 3 270M

 The compact model for hyper-efficient AI

 Read blog July 2025 Introducing T5Gemma

 A new collection of encoder-decoder Gemma models

 Read blog July 2025 Introducing MedGemma

 Our most capable open models for health AI development

 Read blog Introducing Gemma 4

 Frontier-level capabilities and mobile-first AI

 Slide 1 of 2 Your browser does not support the video tag. Your browser does not support the video tag. E2B & E4B Maximum compute and memory efficiency

 A new level of intelligence for mobile and IoT devices

 Learn more Your browser does not support the video tag. Your browser does not support the video tag. 26B & 31B Advanced reasoning

 Frontier intelligence on personal computers

 Learn more Your browser does not support the video tag. Your browser does not support the video tag. E2B & E4B Maximum compute and memory efficiency

 A new level of intelligence for mobile and IoT devices

 Learn more Your browser does not support the video tag. Your browser do

... (truncated, 6 KB total)

Resource ID: 94f1bbbbf04845d5 | Stable ID: MTMyZjJkZj