Gemma 3-270M
Gemma 3-270M is a foundation model developed by Google DeepMind. Despite its compact size, it is equipped with a broad 128K context window.
Parameters
2.0B
Context Window
128K
License
Gemma License
Release Date
2025-08-14
API Pricing
API pricing for this model is not yet available
Strengths
- ・Lightweight and efficient model size
- ・Supports long 128K context
- ・Developed by Google DeepMind
Weaknesses
- ・Limited parameters, unsuitable for complex tasks
- ・Limited knowledge compared to larger models
- ・Lacks advanced reasoning capabilities
Use Cases
- ・Low-latency operation on edge devices
- ・Small-scale text generation tasks
- ・Efficient document context processing
Deep Analysis
Parameters
270M
170M embedding + 100M transformer blocks
Vocabulary
256K tokens
Large vocabulary for handling rare/specific tokens
Release Date
August 14, 2025
Part of Gemma 3 family expansion
Battery Usage
0.75% per 25 conversations
Tested on Pixel 9 Pro SoC with INT4 quantization
Quantization
INT4 QAT
Quantization-Aware Trained checkpoints available
License
Gemma Terms of Use
Gemma Terms of Use
IFEval
Best-in-class for size
New level of instruction-following performance for 270M class
Strengths
- ・Extremely energy efficient: 0.75% battery for 25 conversations on Pixel 9 Pro
- ・Strong instruction-following out of the box (best IFEval for its size class)
- ・256K vocabulary handles rare and domain-specific tokens well
- ・Production-ready INT4 QAT quantization for edge deployment
- ・Excellent base for task-specific fine-tuning
Weaknesses
- ・Not designed for complex conversational use cases
- ・Very limited general knowledge and reasoning capability
- ・Text-only (no multimodal support)
- ・Small context window compared to larger models
- ・Requires fine-tuning for production quality on specific tasks
Competitor Comparison
| Model | Arena | SWE | GPQA | Price |
|---|---|---|---|---|
| Gemma 3 270M | N/A | N/A | N/A | Free (open weights) |
| Phi-3.5 Mini (3.8B) | ~1100 | N/A | ~55% | Free (open weights) |
| SmolLM2 (1.7B) | N/A | N/A | N/A | Free (open weights) |
| Qwen2.5 (0.5B) | N/A | N/A | N/A | Free (open weights) |
Gemma 3 270M is Google's most compact open model, designed for task-specific fine-tuning with strong instruction-following capabilities. With only 270M parameters and extreme energy efficiency (0.75% battery per 25 conversations on Pixel 9 Pro), it targets on-device AI, IoT, and specialized classification/extraction tasks.
Sources
Analysis generated: 2026-05-24