Alibaba has released Qwen3.7-Plus, the latest model in the Qwen family, with multimodal capabilities supporting text, video, and imagery at a cost of $0.40 per million tokens input and $1.60 output. Compared to the predecessor Qwen3.7-Max, the total cost is 60% lower, but the model is only available via proprietary API, marking a shift from Alibaba's open source strategy. This change has implications for enterprises relying on open weights for compliance and data sovereignty.
A model built for enterprise AI agents
Qwen3.7-Plus introduces the preserve_thinking parameter, which maintains internal reasoning across multiple conversational turns, solving the state decay problem in autonomous agents. With a 1-million token context window and 256K tokens dedicated to chain-of-thought, the model is ideal for complex automation tasks such as cloud migration or codebase analysis. In benchmarks, it surpasses GPT-4.5 on ScreenSpot Pro (79.0 vs 67.4) and Terminal Bench 2.0 (70.3 vs 67.9 for DeepSeek-V4-Pro Max).
Google responds with publisher control
Meanwhile, Google announced that websites can opt out of AI-based search results without affecting their placement in regular searches. The move follows pressure from publishers concerned about their content being used in AI-generated summaries powered by Gemini. Google states that opting out will not impact normal ranking, offering a middle ground between transparency and innovation.
Implications for the AI ecosystem
On one hand, Alibaba democratizes access to powerful multimodal models, pushing enterprises to rethink their AI agent architectures. On the other hand, Google recognizes the right of creators to control their visibility in AI Search. Together, these developments reshape the rules for enterprise AI adoption, as already highlighted by the AI budget challenges faced by Uber in 2026. For deeper analysis, see VentureBeat.
Sponsored Protocol
