OpenRouter posted on X that Qwen3.7-Max from Alibaba Qwen is now live on its platform, describing it as the flagship of the Qwen3.7 series for "agent-centric" work such as coding, office and productivity tasks, and long-horizon autonomous execution. The post also claims "big jumps" in coding and agent benchmarks over Qwen3.6, alongside explicit prompt caching for repeated context.
OpenRouter also linked to a try-it page and a separate guide on using explicit caching with Qwen models.
The response from the thread was mostly cautious enthusiasm. One commenter called "agent-centric" the key phrase, while another argued that claims like these should come with more evidence, including recovered tool calls, checkpoint frequency, human interventions, final diff quality, and cost per accepted repo change.
Source: OpenRouter
