

Qwen 3.6
#3alibaba · v3.6 · seit April 2026 · 160× · zuletzt 30. Juni 2026
72
Momentum
Qwen 3.6 ist ein Sprachmodell von Alibaba, das in mehreren Größen verfügbar ist (27B, 35B). Es handelt sich um ein Mixture-of-Experts Agenten-Modell, das mit Multi-Token-Prediction ausgestattet ist und Leistung im GPT-4-Klasse-Bereich erreicht. Das Modell wurde teils als Open-Source-Variante veröffentlicht.
Momentum-Verlauf
04.04.03.07.
Features
| Key-Benchmark (%) | Qwen3.6-35B-A3B: SWE-Bench Verified 73,4 %; Terminal-Bench 2.0: 51,5 %; AIME 2026: 92,6 %. Qwen3.6-Plus: SWE-Bench Verified 78,8 % |
| Kontextfenster (Token) | Qwen3.6-35B-A3B: 262.144 nativ, erweiterbar auf 1.010.000 (via YaRN); Qwen3.6-Plus: 1.000.000 nativ; Qwen3.6-Max-Preview: 256.000 |
| Lizenz | Qwen3.6-35B-A3B: Apache 2.0 (Open Weights, kommerziell nutzbar); Qwen3.6-Plus / Max-Preview: proprietär, API-only |
| Multimodalität | Ja – Text, Bild, Video als Input (nativ multimodal); Qwen3.6-Max-Preview ist text-only (kein Bild-Input) |
| Plattform | Alibaba Cloud Model Studio (API); Qwen Studio (Chat-UI); Hugging Face & ModelScope (Open Weights); OpenRouter, Together AI, DeepInfra (Drittanbieter) |
| Preis | Qwen3.6-Plus: $0,325/1M Input, $1,95/1M Output (OpenRouter); Qwen3.6-35B-A3B: $0,14/1M Input, $1,00/1M Output (OpenRouter) |
| Preis pro 1M Token | Qwen3.6-Plus: $0,325 Input / $1,95 Output; Qwen3.6-35B-A3B: $0,14 Input / $1,00 Output (via OpenRouter/Alibaba Cloud) |
| Release-Datum | Qwen3.6-Plus: 02.04.2026 (API); Qwen3.6-35B-A3B Open Weights: 16.04.2026; Qwen3.6-Max-Preview: 20.04.2026 |