Gemma 3: multimodal, 128K context, 140+ languages
Google's Gemma 3 comes in 1B, 4B, 12B and 27B sizes; the 4B and up accept images, support 128K context and 140+ languages, and the 27B instruct model is reported to beat Gemini 1.5 Pro.
What happened
Google released Gemma 3, with base and instruction-tuned models on the Hub and Hugging Face integration.
The technical details
From the Hugging Face post (comparing with Gemma 2):
| Gemma 2 | Gemma 3 | |
|---|---|---|
| sizes | 2B, 9B, 27B | 1B, 4B, 12B, 27B |
| context | 8K | 32K (1B), 128K (4B, 12B, 27B) |
| images + text | no | yes for 4B, 12B and 27B (1B is text-only) |
| languages | – | English (1B), 140+ (larger sizes) |
The post quotes two headline results: Gemma-3-4B-IT beats Gemma-2-27B-IT, and Gemma-3-27B-IT beats Gemini 1.5-Pro across benchmarks.
Why it matters
A 4B model beating its 27B predecessor is the same story as phi and SmolLM: better data and training let small models catch up fast.
#llm#multimodal#open-weights#google
Sources
This is a summary written from the linked public sources. Figures and numbers are as reported there.