Compare LLM Models
Select 2-5 models to compare side by side.
Select models to compare
0 / 5 selected
Top downloads
Select 2-5 models
| Attribute |
google
|
|---|---|
| Company | |
| Release Date | 2025-06-26 |
| Parameters (B) | 2 |
| Architecture | MatFormer (Matryoshka Transformer) |
| Context Window | 32768 |
| Input Modalities | text, image, audio |
| Open Source | Yes |
| License | gemma |
| Score | 0 |
| VRAM (GB) | 11 |
| Compute | MatFormer E2B variant with ~2B effective parameters; ~11GB VRAM for full bf16 GPU inference (~10.88GB weights), or ~2GB accelerator memory with Per-Layer Embeddings offloaded to CPU for on-device deployment |
| Benchmark: AIME-2025 | 6.7 |
| Benchmark: ARC | 51.7 |
| Benchmark: BBH | 44.3 |
| Benchmark: DROP | 53.9 |
| Benchmark: GPQA-Diamond | 24.8 |
| Benchmark: HellaSwag | 72.2 |
| Benchmark: HumanEval | 66.5 |
| Benchmark: LiveCodeBench | 18.6 |
| Benchmark: MBPP | 56.6 |
| Benchmark: MMLU | 60.1 |
| Benchmark: MMLU-Pro | 40.5 |
| Pricing (Input) | — |
| Pricing (Output) | — |