Models · Compare
OLMo 3 32B Instruct vs Gemini 3 Pro
Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.
Specs
| Field | A: OLMo 3 32B Instruct | B: Gemini 3 Pro |
|---|---|---|
| Released | 2025-11-20 | 2025-11-18 |
| Developer | AI2 | Google DeepMind |
| Openness | Open | Proprietary |
| License | Apache-2.0 | Proprietary |
| OSI-approved | yes | no |
| Data released | yes | no |
| Training code | yes | no |
| Architecture | dense | unknown |
| Total params | 32B | — |
| Active params | — | — |
| Experts | — | — |
| Context window | 66K | 1.0M |
| Attention | gqa | unknown |
| Position enc. | rope | unknown |
| Pretraining tokens | 6.0T | — |
| Post-training | sft, dpo, rlvr | rlhf |
| Training hardware | — | — |
| $/M input | — | $2.00 |
| $/M output | — | $12.00 |
| Output tok/sec | — | 132.9 |
Benchmarks
Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.
General reasoning
| MMLU | 85.4 2025-11-20 | — |
| MMLU-Pro | — | 89.8 2026-05-21 |
| GPQA-Diamond | — | 91.9 2025-11-18 |
Code
| SWE-Bench Verified | — | 76.2 2025-11-18 |
| LiveCodeBench | — | 91.7 2026-05-21 |
Math
| MATH | 96.1 2025-11-20 | — |
| AIME 2024 | 76.8 2025-11-20 | — |
| AIME 2025 | — | 95.7 2026-05-21 |
Held-out / arena
| IFEval | 89.0 2025-11-20 | — |
Context · A
AI2's first 32B fully open frontier model, released November 20 2025 alongside a 7B and Base, Think, Instruct, and RL-Zero variants. Built on the Dolma 3 9.3T-token corpus (5.9T used for pretraining) and the Dolci post-training suite. All weights, data descriptions, intermediate checkpoints, and code released.
Context · B
DeepMind's flagship Gemini 3, announced by Sundar Pichai November 18 2025 with broader rollout on November 20. Shipped with a Deep Think mode and an agentic Gemini Agent surface, plus the new Antigravity platform for agent-first development.