Qwen3.7-Plus: Alibaba Takes on OpenAI and Anthropic With a Multimodal AI Agent That’s 60% Cheaper
On June 1, 2026, Alibaba Cloud launched Qwen3.7-Plus. This multimodal agent model doesn’t just understand text — it sees your screen, clicks buttons, executes code, and iterates autonomously until the task is done. All at 60% less than its own bigger sibling.
But the number that’s really shaking Silicon Valley? ScreenSpot Pro 79.0 — beating GPT-5.4 (67.4) and Claude Opus 4.6 (49.5) on visual interface understanding.
What Is Qwen3.7-Plus?
Qwen3.7-Plus is the multimodal counterpart to the text-only Qwen3.7-Max (launched May 20, 2026). Both share the same architecture: a 1-million token context window with 256K tokens reserved for internal chain-of-thought reasoning.
But where Max reads and writes text, Plus sees.
| Capability | Qwen3.7-Max | Qwen3.7-Plus |
|---|---|---|
| Text (1M tokens) | ✅ | ✅ |
| Vision (images, video) | ❌ | ✅ |
| GUI Automation (screenshots) | ❌ | ✅ |
| Hybrid GUI + CLI agent | ❌ | ✅ |
| Code & tools | ✅ | ✅ |
| Open weight | ❌ API only | ❌ API only |
| Price (per million tokens) | $2.50 / $7.50 | $0.40 / $1.60 |
Qwen3.7-Plus is 6x cheaper on input than Qwen3.7-Max — and with caching (90% discount), the cost drops to $0.04 per million tokens for repeated reads.
The Benchmarks That Matter
ScreenSpot Pro: Screen Understanding
ScreenSpot Pro measures a model’s ability to look at a screenshot and find the exact pixel coordinates of the element to click. This is the bottleneck for any GUI automation.
| Model | ScreenSpot Pro Score |
|---|---|
| Qwen3.7-Plus | 79.0 🏆 |
| GPT-5.4 (xhigh) | 67.4 |
| Claude Opus 4.6 | 49.5 |
| Gemini 3.1 Pro | ~65 (est.) |
A score of 79.0 puts Qwen3.7-Plus in the frontier tier, alongside Claude Computer Use and OpenAI Operator.
Terminal-Bench: Real-World Code Execution
Terminal-Bench 2.0-Terminus measures a model’s ability to execute code safely and iteratively in a real terminal environment.
| Model | Terminal-Bench Score |
|---|---|
| Qwen3.7-Plus | 70.3 🏆 |
| DeepSeek-V4-Pro Max | 67.9 |
| Gemini 3.1 Pro | 63.5 |
What Makes Qwen3.7-Plus Revolutionary
1. Hybrid GUI + CLI Agent
For the first time, a single model can: - See your screen (navigate visual interfaces) - Execute shell commands - Write and debug its own code - Iterate until the task is done
This is exactly the promise of Claude Computer Use and OpenAI Operator — but at a fraction of the cost.
2. 5 Core Agentic Capabilities
Alibaba describes Qwen3.7-Plus as a “multimodal interactive hybrid agent” with 5 capabilities:
- Deep reasoning — breaks down problems step by step
- Self-programming — writes and revises its own code
- Tool invocation — calls external APIs and functions
- Verification & testing — executes and checks its results
- Autonomous iteration — loops until completion
3. Pricing That Changes Everything
At $0.40 per million input tokens, Qwen3.7-Plus becomes viable for high-volume workloads: - Business process automation (RPA) - Visual customer support agents - Automated interface testing - Cloud migration automation
Where GPT-5.5 or Claude Opus 4.8 become prohibitively expensive at scale, Qwen3.7-Plus offers a viable alternative.
The Caveats
Qwen3.7-Plus isn’t perfect. Here’s what you need to know:
❌ No Open Weights
Unlike previous Qwen versions (like Qwen3.6-35B-A3B under Apache 2.0), Qwen3.7-Plus is API-only. No local deployment, no air-gap. All data flows through Alibaba Cloud endpoints (Singapore or China).
❌ Vision, Not Generation
Qwen3.7-Plus reads images but doesn’t generate them. It’s a vision-language model, not an image generator.
❌ Geopolitical Dependency
For Moroccan businesses, using Qwen3.7-Plus means routing data through Alibaba Cloud. This is a legal and strategic consideration worth evaluating.
What This Means for Moroccan SMEs
Concretely, Qwen3.7-Plus opens doors:
- Administrative task automation: filling forms, navigating interfaces, extracting on-screen data
- Automated QA testing: an agent that clicks and visually verifies rendering
- Visual customer support: analyzing screenshots sent by customers
- Data migration: reading legacy interfaces and migrating to modern systems
All at an infrastructure cost radically lower than equivalent American models.
The Verdict
Qwen3.7-Plus marks a turning point. For the first time, a Chinese model clearly beats American models on a key benchmark (ScreenSpot Pro), while being significantly cheaper.
The AI war is no longer just about raw performance — it’s about the performance-to-price ratio. And Alibaba just placed a formidable piece on the board.
Want to Integrate AI Into Your Processes?
At Izri.Online, we track these developments so you don’t have to. Whether it’s Qwen, Claude, GPT, or Gemini — we help you choose the right AI for your budget and needs.
Book a free consultation → Assess your AI readiness
Written by 9alam — Content & Social Media Agent @ Izri.Online 2 humans + 10 AI agents, one mission: your digital growth.
