Google Gemma 3 27B: largest Gemma 3 open model, most capable on a single GPU or TPU, 14T training tokens, SigLIP vision, 128K context, 140+ languages.
Capabilities, design details, and architectural traits
Gemma 3 27B is the largest model in Google's Gemma 3 family and is officially described as the most capable model that fits on a single GPU or TPU host. It is a multimodal open-weights model, accepting text and image input and generating text output. It was trained on 14 trillion tokens - the largest training budget in the Gemma 3 family - and supports a 128K context window with multilingual coverage across over 140 languages.
| Trait | Detail |
|---|---|
| Single GPU/TPU fit | Documented as the most capable open model deployable on a single consumer GPU or TPU |
| Training tokens | 14 trillion - largest in the Gemma 3 family (vs 12T for 12B, 4T for 4B) |
| SigLIP vision encoder | Shared frozen SigLIP encoder (same across 4B, 12B, and 27B); processes images as 256 compact soft tokens via MultiModalProjector |
| Pan and Scan image handling | Adaptively crops non-square or high-resolution images into 896x896 tiles for improved document and text-in-image tasks |
| Context window | 128K tokens; achieved by scaling from 32K pre-training via RoPE rescaling with a base frequency of 1M on global attention layers |
The Gemma 3 27B multimodal architecture serves as the base model for MedGemma 27B, Google's open medical AI model designed for health AI development. This makes the 27B size the specific Gemma 3 variant chosen for specialized downstream medical fine-tuning in Google's own research ecosystem.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | |
| Release Date | March 12, 2025 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | Aug 2024 | May 2026 | - |
| Context & Limits | |||
| Context Window | 131K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $5 | $10 |
| Output Pricing | Free Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | textimage | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 7.4 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | 10.1 | 78.0 Best Coding Index | 76.5 |
| Agentic Index | 0.3 | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
