Qwen: Qwen3.7 Plus
Multimodal agent model from Alibaba Cloud scoring 79.0 on ScreenSpot Pro and 81.0 on AndroidWorld, with a ~977K token context window and pricing at $0.40 input / $1.60 output per million tokens. Offers strong GUI agent and long-context RAG capabilities at a fraction of the cost of frontier-tier peers.
Specifications
| Attribute | Value |
|---|---|
| Lab | Qwen |
| Tags | Intelligent Image Input Long Context |
| Release Date | 2026-06 |
| Context Window | 1,000,000 tokens |
| Input Price / 1M | $0.40 |
| Output Price / 1M | $1.60 |
| Input Modalities | Text, Image |
| Output Modalities | Text |
Strengths
- 79.0 on ScreenSpot Pro (GUI agent benchmark)
- 81.0 on AndroidWorld (mobile agent task score)
- ~977K token context window for long-document RAG
- Multimodal: text, image, and video understanding
- Low pricing at $0.40 input per million tokens
Weaknesses
- Slower output speed (~49.8 tok/s) vs peers
- High TTFT latency of 3.26 seconds
- Not open-weight; API-only access
- Lags behind Alibaba's own Qwen3.7-Max on coding
Best For
- GUI agent automation (ScreenSpot Pro 79.0)
- Long-context RAG with ~977K tokens
- Vision-to-code from screenshots or video
- Autonomous agent loops at low cost
In Depth: Qwen3.7 Plus
Benchmark Performance
Qwen: Qwen3.7 Plus , placing #38 overall. Its agent-specific benchmarks 79.0 on ScreenSpot Pro and 81.0 on AndroidWorld exceed its general rank, outperforming Claude Sonnet 5 (LMRank 9.4/10, #6) on GUI tasks while costing 5x less per output token.
On vendor-reported tests, Qwen: Qwen3.7 Plus achieves 90.3 on GPQA Diamond and 89.6 on MMLU-Pro. Its agentic focus is validated by ScreenSpot Pro 79.0 and AndroidWorld 81.0. The independent Artificial Analysis Intelligence Index scores it at 39, well above the median of 16. These results show strong multimodal reasoning for a model priced 40x below GPT-5.5 Pro ($30 in / $180 out).
Pricing & Value
Qwen: Qwen3.7 Plus costs $0.40 per million input tokens and $1.60 per million output tokens directly, or $0.32 / $1.28 on OpenRouter. That is 37x cheaper than Claude Fable 5 ($10 / $50) for input and 31x cheaper for output.
At $0.40 input / $1.60 output, Qwen: Qwen3.7 Plus delivers 0.24 LMRank points per dollar of input (8.5 points ÷ $34.00 round-trip cost for 1M tokens). By comparison, GPT-5.3 ($11.00 in / $44.00 out, LMRank 9.4) yields 0.17 points per dollar. For developers running high-volume agent loops, Qwen3.7 Plus offers 1.4x the value-per-point at 1/27th the absolute cost of GPT-5.5 Pro.
Who Should Use This
GUI agent developers needing 79.0 ScreenSpot Pro scores on a budget. RAG engineers processing ~977K-token documents. Vision-to-code teams converting screenshots or video to working code.
- GUI agent developers: 79.0 ScreenSpot Pro at 1/31st Claude Fable 5's output cost
- Long-context RAG builders: ~977K tokens at $0.40/M input with no hidden fees
- Vision-to-code teams: multimodal input for screenshot-driven code generation
- Anti-persona: Not for speed-critical production 3.26s TTFT and 49.8 tok/s limit throughput
Release & Version History
Qwen: Qwen3.7 Plus launched in June 2026 as Alibaba Cloud's multimodal agent extension of the text-only Qwen3.7 backbone, sitting below the higher-cost Qwen3.7-Max flagship.
Qwen3.7 Plus is the multimodal sibling to Qwen3.7 (text-only). It succeeds Qwen3.6-Plus from early 2026, adding video understanding and a ~977K context window. Unlike the open-weight Qwen family tradition, this model is proprietary and API-only, available via Alibaba Cloud Model Studio and OpenRouter. It sets a budget agentic tier while Qwen3.7-Max targets top-tier coding benchmarks.