Alibaba’s Qwen team has unveiled Qwen3.6-Max-Preview, describing it as an early look at its next flagship model with a focus on agentic coding, instruction following, and “real-world” reliability.
In its announcement, Qwen positioned Max-Preview as an upgrade over Qwen3.6-Plus, with “improved agentic coding capability,” “stronger world knowledge and instruction following,” and better “real-world agent and knowledge reliability performance.” The company also said “more Qwen3.6 models” are on the way.
What Qwen is emphasizing this round
The headline feature is agentic coding, and responses quickly honed in on whether the gains are about writing code—or handling longer, more fragile workflows.
Early users claimed the latest update shows notable benchmark deltas over 3.6-Plus—“SkillsBench +9.9 and Terminal-Bench 2.0 +3.8”—and framed Max-Preview as a proprietary tier available through “Model Studio,” contrasting it with open-weight releases like “3.6-35B-A3B.”
The questions users are already asking
Pricing also surfaced. Users cited token pricing for “Qwen3-Max” on Alibaba Cloud and compared it to Claude Opus pricing, arguing the spread functions as a market signal for teams running coding agents in production. ATOMInference similarly said Qwen is “consistently one of the most cost-competitive model families,” while watching where Max-Preview lands.
Source: https://x.com/Alibaba_Qwen/status/2046227759475921291