Models · Compare
Gemma 3 27B IT vs Gemini 2.5 Pro
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: Gemma 3 27B IT | B: Gemini 2.5 Pro |
|---|---|---|
| Released | 2025-03-12 | 2025-03-25 |
| Developer | Google DeepMind | Google DeepMind |
| Openness | Source-available | Proprietary |
| License | Gemma Terms of Use | Proprietary |
| OSI-approved | no | no |
| Data released | no | no |
| Training code | no | no |
| Architecture | dense | unknown |
| Total params | 27B | — |
| Active params | — | — |
| Experts | — | — |
| Context window | 131K | 1.0M |
| Attention | hybrid-gqa-sliding | unknown |
| Position enc. | rope | unknown |
| Pretraining tokens | 14.0T | — |
| Post-training | sft, rlhf | rlhf |
| Training hardware | TPU v5p | — |
| $/M input | — | $1.25 |
| $/M output | — | $10.00 |
| Output tok/sec | — | 127.3 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU-Pro | 67.5 2025-03-12 | 86.2 2026-05-21 |
Code
| HumanEval | 87.8 2025-03-12 | — |
| LiveCodeBench | — | 80.1 2026-05-21 |
Math
| MATH | — | 96.7 2026-05-21 |
| AIME 2025 | — | 87.7 2026-05-21 |
Held-out / arena
Context · A
Gemma 3 27B IT introduced native multimodal input (text plus images at 896x896 normalized to 256 tokens) and pushed context to 128K at the 27B scale. Trained on 14T tokens across 140+ languages on TPU v5p. The Gemma Terms of Use remain source-available rather than OSI-approved.
Context · B
The first Gemini release to clearly lead on LMArena Elo and on hard reasoning benchmarks. Native 1M-token context with reported 2M expansion in pipeline.