SummaryAlibaba has made its 2.4-trillion-parameter Qwen4-Max model generally available. The model offers a 1-million-token context window, native visual input support, and API pricing that reduces average task costs by approximately 58% according to its own benchmarks. |
Why it mattersThis aggressive price cut arrived on the same day Tencent announced a 95% reduction in internal costs. This indicates a coordinated move by Chinese companies to undercut US leading model prices, making large-context and multimodal inference economically viable for a broad developer base. Plans to release smaller Qwen4 variants as open source could further lower the cost floor for the entire ecosystem. |
Highlights- 2.4-trillion-parameter MoE, 1-million-token context window, and direct visual input
- Approximately 58% reduction in average task cost in Alibaba's Work360 benchmark (26.58 RMB → 11.16 RMB)
- Plans to release smaller Qwen4 models as open source
|
Critical takeThe 58% cost savings cited in the headline were measured solely in Alibaba's own Work360 benchmark. Since no independent or cross-provider evaluation is referenced, it is difficult to verify this figure or compare it with competitors' claims. The article also lacks third-party data on quality or latency, leaving the true capability-per-yuan balance untested. |
This critique was written by the local AI model
Qwen3.8-27B. I chose the story; I do not review each critique before it is published.