Alibaba Unveils Multimodal Model Qwen3.7-Plus with Low Token Costs

Alibaba Cloud has announced Qwen3.7-Plus, an upgraded version of its language model that now processes not only text but also video and images. The model is positioned as a cost-effective solution for enterprise tasks, including visual content generation and multimedia data analysis. Usage costs are set at $0.4 per 1 million input tokens and $1.6 per 1 million output tokens, making it one of the most affordable powerful AI models on the market.

The new model features a context window of 1 million tokens, with up to 256,000 tokens allocated for internal reasoning chains. This enables the system to maintain logical consistency when performing multi-step tasks such as cloud infrastructure migration or code dependency analysis. A key feature is the preserve_thinking parameter, which retains internal reasoning blocks between queries to prevent context loss. Similar mechanisms are used by Anthropic (Extended Thinking) and OpenAI (encrypted reasoning pass-back).

In benchmarks, Qwen3.7-Plus delivers competitive results. On the Terminal Bench 2.0-Terminus test, which evaluates safe code execution, the model scored 70.3, outperforming DeepSeek-V4-Pro Max (67.9) and Gemini-3.1 Pro (63.5). In computer vision tasks like ScreenSpot Pro, it achieved 79.0, surpassing GPT-5.4 (67.4) and Claude Opus 4.6 (49.5). However, in overall capabilities, the model still lags behind flagship offerings from U.S. companies.

For enterprise users, Qwen3.7-Plus could serve as a replacement for expensive models in automation, process robotization, and data processing tasks. The model’s API is OpenAI-compatible, simplifying integration into existing systems. Alibaba also offers flexible pricing: cached data is processed at $0.04 per 1 million tokens, reducing costs for repetitive operations. However, the lack of open weights and the requirement to use cloud APIs may pose limitations for companies with strict data security and localization requirements.