Side-by-side comparison of Gemma 4 26B A4B Instruct (Google DeepMind · United States) and Qwen3.6-27B (Alibaba (Qwen)) for self-hosted deployment of the open-weight model. Gemma 4 26B A4B Instruct is rated conditional; Qwen3.6-27B is conditional. They part ways on training data: Gemma 4 26B A4B Instruct is "Domain-level summary", Qwen3.6-27B is "Undisclosed".
| 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 Per the published Apache 2.0 licence, the Qwen3.6-27B weights are deployable without commercial restriction, including for vision-language and 1M-context workloads. The blockers for regulated EU use are the China-based vendor and the absence of any training-data disclosure on the model card — both should be mitigated through self-hosting and a deployer-prepared GPAI compliance file. |
| Last reviewed | 2026-04-17 | 2026-04-28 |
| Open-weight | ||
| Licence | Apache 2.0 | Apache 2.0 |
| Commercial use | Unrestricted | Unrestricted |
| Training data | Domain-level summary | Undisclosed |
| Origin | United States | China |
| Performance & pricing? | ||
| Quality index | 27/100 | 46/100 |
| Speed | — | 66 tok/s |
| Blended price | — | $1.35/M |
| Context window | — | — |
| Evidence | ||
| Sources | ||
No overlapping sources between the two entries.