Qwen: Qwen3.6 Plus
Alibaba
Capabilities
Strengths
- Excellent long-context handling (1M token window)
- Strong multimodal performance (text, image, video)
- Extended thinking and structured output support
- Open weights enable self-hosting and full privacy control
Weaknesses
- Very slow inference (52 tok/s; flagship-level latency cost)
- Sparse reasoning/analysis benchmarks vs. larger peers
- Training data transparency limited for open-weight model
- Hybrid architecture complexity may hinder deployment optimization
Pricing
Input / 1M tokens
$0.325
Output / 1M tokens
$1.95
Context window
1,000k
Transparency
Open weights
10.0 / 10
Open training data
0.0 / 10
Open methodology
7.2 / 10
Licence openness
8.5 / 10
Provider disclosure
6.8 / 10
FMTI company score
N/A
Open-weight model with published architecture details (linear attention + sparse MoE hybrid). Training data composition not fully disclosed. Alibaba publishes moderate technical methodology; less transparency than Meta but more than most closed providers.
Sustainability
Inference energy
N/A
Training footprint
N/A
Provider infrastructure
5.5 / 10
Hybrid architecture (linear attention + sparse MoE) suggests efficiency gains over dense models. Alibaba cloud infrastructure sustainability not well-documented publicly. Conservative estimate pending detailed energy data.