Model comparison
Gemma 4 26B-A4B Instruct (MoE) vs StarCoder2 15B
Capability radar, value bars, and side-by-side posture.
Gemma 4 26B-A4B Instruct (MoE)
Not scored
value score
StarCoder2 15B
70
value score
Gemma 4 26B-A4B Instruct (MoE) ctx
—
StarCoder2 15B ctx
16K
Capability radar
Value · context · multimodal · openness · speed posture
Value head-to-head
- StarCoder2 15B70
Attribute tape
| Field | Gemma 4 26B-A4B Instruct (MoE) | StarCoder2 15B |
|---|---|---|
| Developer | Hugging Face | Hugging Face |
| Context | — | 16K |
| Modalities | Image-text-to-text | Text, Code |
| Openness | — | Open Weights |
| Speed | — | — |
| Price | — | — |
| Value score | — | 70 |
| Summary | Open-weight multimodal instruction-tuned model from Google DeepMind, built on a Mixture-of-Experts architecture with ~25.2B total parameters and ~3.8B active per token, designed for long-context reasoning, vision-language tasks, function calling, and tool-using agents. | BigCode open coding model on Hugging Face. |