Alibaba's Qwen3.7-Plus sees screens, codes apps, beats GPT-5.4 in GUI benchmarks

Alibaba's Qwen team has launched Qwen3.7-Plus, a multimodal agent model that combines visual perception, graphical user interface control, and code generation within a single autonomous agent loop. The model accepts text, images, and video as input but outputs only text, enabling it to read screens, navigate apps, write code from visual templates, and invoke external tools without human intervention. In demonstrations, it autonomously built an English vocabulary app over 11 hours, generating over 10,000 lines of code, and recreated Apple's Stocks app by parsing UI structure from screenshots. On Alibaba's vendor-stated benchmarks, Qwen3.7-Plus leads on GUI-grounding tasks: 79.0 on ScreenSpot Pro (vs GPT-5.4's 67.4) and 81.0 on AndroidWorld (vs Gemini 3.1 Pro's 70.7). The model supports a 1-million-token context window, up to 65,536 output tokens, and an internal chain-of-thought reasoning budget of up to 256,000 tokens. Pricing is aggressive at $0.40 per million input tokens and $1.60 per million output tokens, roughly 6x cheaper on input than the text-only Qwen3.7-Max. Cached input is as low as $0.04 per million tokens.

A key technical feature is the 'preserve_thinking' parameter, which maintains the model's reasoning state across multiple tool calls in an agent loop, preventing reset of its chain of thought. The model also supports the Anthropic API protocol and integrates with Claude Code and third-party coding tools. However, independent evaluations from Artificial Analysis rank it #53 out of 164 on the Intelligence Index, with notably slow output speed (~52.9 tokens per second) and verbosity issues (generating 110M output tokens vs 29M median). On pure reasoning tasks like GPQA Diamond, it trails Claude Opus 4.6 Max (90.3 vs 91.3), and on SWE-Bench Pro it scores below its text-only sibling Qwen3.7-Max. The model is proprietary and API-only, marking a shift from Alibaba's previous open-source Qwen releases. Alibaba shares rose up to 6.84% on the announcement as the company positions itself against OpenAI, Anthropic, and Google in the emerging autonomous agent market.