Alibaba releases Qwen4-Max, a 2.4T-parameter MoE model
Start here
Alibaba's Qwen team released Qwen4-Max, the first model in its Qwen4 generation: a proprietary sparse MoE model with a 2.4T-parameter backbone, 71B active parameters per token, text, image and video input, and a 1M-token context window. It is hosted only, through QwenCloud API, Token Plan, Qwen Studio, QwenWork and Qoder, with no open weights, so it cannot be self-hosted or run as quantized builds; a Token Plan-only preview opened this month, and the default rate limit is 5,000,000 tokens per minute. Alibaba reports a 62% drop in average cost per task, with input priced at $1.50 and output at $4.50 per million tokens, down from $2 and $6 for Qwen3.8-Max. Qwen4-Max ranks first on 4 of 16 charted benchmarks, led by HLAB-Vals-1.1 at 21.3, but trails clearly on coding and vision, scoring 31.5 on ProgramBench versus 65.0 for Claude Opus 5.5.
Generated from reports · updated 13 min ago
Report timeline
Follow the reports to see the event from different sides.
- MarkTechPostPicksAlibaba releases Qwen4-Max: 2.4T MoE with 71B active parameters and 1M context
Alibaba's Qwen team released Qwen4-Max, the first model in its Qwen4 generation, a proprietary sparse MoE model with a 2.4T-parameter backbone, 71B active parameters per token, text, image and video input, and a 1M-token context window. Alibaba reports a 62% drop in average cost per task, with input priced at $1.50 and output at $4.50 per million tokens, down from $2 and $6 for Qwen3.8-Max. Qwen4-Max ranks first on 4 of 16 charted benchmarks, led by HLAB-Vals-1.1 at 21.3, but trails clearly on coding and vision, scoring 31.5 on ProgramBench versus 65.0 for Claude Opus 5.5.
Heat for this event
Not enough continuous observations yet to draw a trend.