Side-by-side comparison of Gemma 4 26B A4B Instruct (Google DeepMind · United States) and QwQ-32B (Alibaba · China) for self-hosted deployment of the open-weight model. Gemma 4 26B A4B Instruct is rated conditional; QwQ-32B is conditional. They part ways on commercial use: Gemma 4 26B A4B Instruct is "Unrestricted", QwQ-32B is "Yes".
| Field | ||
|---|---|---|
| Summary | ||
| Verdict | Conditional Based on published licence terms, Gemma 4 26B A4B ships under pure Apache 2.0 with no prohibited-use carve-outs — a departure from prior Gemma generations. The sparse-MoE architecture (25.2B total / 3.8B active) puts it in an ambiguous zone for EU AI Act GPAI systemic-risk classification, and US origin plus image-input support add transparency obligations that deployers should document. | Conditional 32B dense reasoning model under Apache 2.0. Sweet spot for self-hostable reasoning: 4090-class GPU at 4-bit, single H100 at bf16. Chinese-origin caveats unchanged. |
| Last reviewed | 2026-04-17 | 2026-04-15 |
| Open-weight | ||
| Licence | Apache 2.0 | Apache 2.0 |
| Commercial use | Unrestricted | Yes |
| Training data | Domain-level summary | Undisclosed |
| Origin | United States | China |
| Performance & pricing? | ||
| Quality index | 27/100 | 20/100 |
| Speed | — | 33 tok/s |
| Blended price | — | $0.74/M |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.