Alibaba Cloud’s Qwen team has published a dedicated blog post for Qwen3.7-Plus, the vision-capable multimodal counterpart to the text-only Qwen3.7-Max in the 3.7 generation. While Qwen3.7-Max handles text-in, text-out reasoning with a 1M-token context window optimized for long-horizon agent tasks, Qwen3.7-Plus adds native image input — enabling visual question answering, chart and document reasoning, and multimodal agent workflows. The Plus-Preview variant appeared on LM Arena’s Vision leaderboard at #16 overall, a result that placed Alibaba at #5 globally among AI labs in vision preference evaluation and first among Chinese labs; both rankings come from the neutral Arena leaderboard. Like its Max sibling, Qwen3.7-Plus is a closed-weight, API-only model accessible via Alibaba Cloud Model Studio, OpenRouter, and compatible third-party platforms; it supports both the OpenAI and Anthropic API specifications, allowing teams to drop it into existing Claude Code or GPT-compatible toolchains with minimal changes.
The 3.7 generation was first announced at the Alibaba Cloud Summit in Hangzhou on May 20, 2026, alongside Alibaba’s in-house Zhenwu M890 AI accelerator chip (144 GB on-chip memory, 800 GB/s inter-chip bandwidth). Qwen3.7-Max scored 56.6 on the Artificial Analysis Intelligence Index v4.0 — ranked #5 globally, #1 among Chinese models — with Alibaba reporting 35+ hour autonomous agent runs and 1,000+ tool calls per session, though those figures are vendor-reported and not yet third-party verified. No open-weight Qwen3.7 models have been released as of publication; Qwen3.6-35B-A3B under the Apache 2.0 license remains the latest downloadable Qwen checkpoint on Hugging Face. Based on Alibaba’s prior cadence with the 3.6 generation, open-weight mid-tier 3.7 variants are expected by mid-2026.