Qwen 3.7 Max
Qwen3.7-Max is Alibaba’s flagship model in the Qwen3.7 series, designed for agentic, text-based workflows. It excels at coding, debugging, office automation, productivity tasks, tool use, and long-horizon autonomous execution. With a 1-million-token context window and support for outputs of up to 64K tokens, it is ideal for processing large documents, repository-scale coding, multi-step planning, structured content generation, and complex workflows requiring sustained reasoning across hundreds or even thousands of steps.
Related Models
Qwen Image 3.0
Qwen Image 3.0 is Alibaba's next-generation open-weights AI image generation and editing model built on a Diffusion Transformer (DiT) architecture. Supporting both text-to-image (T2I) and image-to-image instruction editing (I2I), it excels at rich content rendering, hyper-realistic detail, and deep contextual understanding. It achieves industry-leading text-in-image typography, rendering clear small text down to 10px across 12 native languages and 20+ fonts for complex multi-column layouts, UI designs, and graphic documents. With open weights available for self-hosting, Qwen Image 3.0 provides highly cost-effective, high-precision visual generation for localized advertising, product UI prototyping, e-commerce graphics, and creative publishing.
Qwen Image 3.0 Pro
Qwen Image 3.0 Pro is Alibaba's flagship third-generation AI image generation model built on a Diffusion Transformer architecture. Engineered for high information density and professional production, it accepts ultra-long prompts of up to 4,500 tokens, allowing a single pass to generate complex layouts like newspaper front pages, multi-panel storyboards, academic papers, and detailed UI interfaces. It delivers micro-level detail and industry-leading typography, rendering legible small text down to 10px across 12 native languages and 20+ fonts.
Qwen 3.8 Max
Qwen3.8-Max is Alibaba’s flagship model in the Qwen3.8 series, designed for agentic, text-based workflows. It excels at coding, debugging, office automation, productivity tasks, tool use, and long-horizon autonomous execution. With a 1-million-token context window and support for outputs of up to 64K tokens, it is ideal for processing large documents, repository-scale coding, multi-step planning, structured content generation, and complex workflows requiring sustained reasoning across hundreds or even thousands of steps.
Qwen 3.5 Omni Plus
Qwen 3.5 Omni Plus is Alibaba Cloud's next-generation native multimodal (Omni) flagship model. Powered by an end-to-end unified architecture, it supports real-time full-duplex interaction across text, image, audio, and video inputs and outputs. It features ultra-low latency streaming inference, expressive speech synthesis, joint vision-audio reasoning, and strong tool-use capability—ideal for real-time voice assistants, AI customer service, video interaction, and digital human agents.
Read Me
Qwen3.7 Max API
Qwen3.7-Max is Alibaba's highest-capability Qwen3.7 model for long-horizon agents, demanding reasoning, and repository-scale software work. Its 1M-token context window, 64K maximum output, and expanded thinking budget give the model room to preserve plans, evidence, code, and tool results across extended tasks.
On iCreat, teams can evaluate Qwen3.7-Max in the Playground and move the same model into production through an OpenAI-compatible Chat Completions API. The qwen3.7-max endpoint supports streaming, thinking control, reasoning effort, function calling, and context caching without requiring a separate integration path.
Model Positioning
Qwen3.7-Max is the reasoning-first flagship of the Qwen3.7 family. It is intended for tasks where correctness, planning depth, and sustained execution matter more than minimizing cost per request.
Unlike Qwen3.7-Plus, which is the balanced multimodal choice, the current qwen3.7-max alias is a text model. Its strongest fit is long-form coding, research synthesis, complex professional analysis, and agents that must coordinate many dependent steps.
Key Features
Long-horizon agent execution. Qwen3.7-Max is designed to maintain objectives, intermediate state, and verification criteria through extended tool-driven workflows.
Frontier-tier reasoning. The Max tier prioritizes difficult coding, mathematical reasoning, architecture decisions, and complex analysis where additional inference can materially improve the result.
Million-token working context. A 1M-token window can hold large repositories, long document sets, tool traces, and accumulated task history, while the 64K maximum output supports substantial deliverables.
Hybrid thinking. Applications can enable thinking for difficult work or disable it for direct responses. The model also exposes a thinking budget larger than the maximum visible answer length.
Agent and tool support. Function calling and built-in tool support make the model suitable for search, code execution, database operations, and multi-system orchestration.
Stable long-context reuse. Context caching can reduce repeated processing when related requests share the same instructions, schemas, reference material, or repository context.
Pricing
| Token Type | Price |
|---|---|
| Input | $2.50 per 1M tokens |
| Output | $7.50 per 1M tokens |
| Cache Read | $0.25 per 1M tokens |
| Cache Write | $3.13 per 1M tokens |
Use Cases
Repository-scale engineering. Analyze large codebases, plan cross-file changes, implement features, investigate regressions, and review the final diff against explicit requirements.
Long-running coding agents. Coordinate terminal commands, tests, documentation, issue context, and repair loops while preserving the original objective across many steps.
Complex professional analysis. Reconcile evidence from contracts, technical reports, policies, or research materials and produce a decision-ready analysis with traceable assumptions.
Agent planning and orchestration. Decompose broad objectives into dependencies, select tools, track progress, recover from failed steps, and verify completion criteria.
Technical design and migration. Compare architectures, identify operational constraints, plan phased migrations, and generate implementation guidance for large systems.
High-difficulty reasoning. Handle mathematical, algorithmic, or logic-intensive tasks where direct generation is less reliable than deliberate analysis.
Model Comparison
Qwen3.7-Max vs Qwen3.6-Max-Preview
| Factor | Qwen3.7-Max | Qwen3.6-Max-Preview |
|---|---|---|
| Positioning | Flagship long-horizon agent model | Previous flagship preview |
| Context Window | 1M tokens | 256K tokens |
| Maximum Output | 64K tokens | 64K tokens |
| Official Input Modalities | Text | Text |
| Reasoning Control | Hybrid thinking; enabled or disabled | Hybrid thinking; enabled or disabled |
| Core Difference | Larger context, higher thinking budget, and stronger sustained agent execution | Shorter context and earlier agent behavior |
| Best Fit | New long-running agents and repository-scale workflows | Existing evaluations pinned to preview behavior |
Qwen3.7-Max vs Claude Opus 4.7 and GPT-5.5
| Factor | Qwen3.7-Max | Claude Opus 4.7 | GPT-5.5 |
|---|---|---|---|
| Positioning | Long-horizon agents and strongest Qwen reasoning | Strict execution and verification for difficult engineering | Agentic coding and professional work |
| Context Window | 1M tokens | 1M tokens | 1.05M tokens |
| Maximum Output | 64K tokens | 128K tokens | 128K tokens |
| Official Input Modalities | Text | Text and image | Text and image |
| Reasoning Control | Hybrid thinking; enabled or disabled | Optional Adaptive Thinking; low to max, including xhigh |
Reasoning effort; none to xhigh |
| Weight Availability | Closed weights | Closed weights | Closed weights |
| Core Difference | Text-first agent foundation with deep long-horizon planning | Longer output plus instruction-heavy coding and review | Longer output plus OpenAI-native agent and computer workflows |
| Best Fit | Large text contexts, autonomous coding, and tool orchestration | Quality-sensitive engineering with explicit validation | Professional agents built around the OpenAI ecosystem |
Why Choose Qwen3.7-Max?
Choose Qwen3.7-Max when the task is expensive to restart: a large repository review, a multi-stage investigation, an agent run with many tool results, or a decision that depends on reconciling extensive evidence. Its value comes from maintaining a coherent plan across a large working context.
The Max tier is less suitable for short classification, simple rewriting, or image and video analysis. For mixed multimodal traffic or routine requests, Qwen3.7-Plus may provide a better capability-cost balance.
Specifications
| Category | Description |
|---|---|
| Model Name | Qwen3.7-Max |
| Provider | Alibaba / Qwen |
| Model ID | qwen3.7-max |
| Official Snapshot ID | qwen3.7-max-2026-05-20 |
| Release Date | May 20, 2026 |
| Model Type | Proprietary text hybrid-thinking model |
| Context Window | 1M tokens |
| Maximum Output | 65,536 tokens |
| Knowledge Cutoff | Not published |
| Official Input Modalities | Text |
| Output Modalities | Text |
| Official Reasoning Control | Hybrid thinking; enabled or disabled per request |
| Total Parameters | Not published |
| Weight Availability | Closed weights |
| iCreat Input Modalities | Text |
| iCreat API Features | OpenAI-compatible Chat Completions API, streaming, hybrid thinking, reasoning effort, function calling, and context caching |
| Best Suited For | Long-horizon agents, repository-scale coding, complex analysis, and tool orchestration |
Architecture
Alibaba has not published the parameter count or detailed network architecture of Qwen3.7-Max. Official materials describe it as a proprietary agent foundation optimized for long-horizon reasoning, coding, tool use, and compatibility with different agent frameworks.
The architecture disclosed for earlier Qwen Max generations should not be presented as a confirmed Qwen3.7-Max specification. In particular, earlier parameter counts or mixture-of-experts details do not establish the configuration of this snapshot.
Notes
- Use thinking for debugging, planning, architecture work, mathematics, and tasks with dependent steps; disable it when a short direct response is sufficient.
- Stream long generations so users can see progress and the client can handle partial output without waiting for the complete response.
- Treat the context window as a total working budget. Remove duplicated logs, generated artifacts, and obsolete task history before each long agent run.
- Put stable instructions, schemas, and shared reference material at the beginning of related requests when using context caching.
- Require explicit checks before destructive commands, deployments, purchases, or external messages; model reasoning is not authorization.
FAQ
Should I choose Qwen3.7-Max or Qwen3.7-Plus?
Choose Max for the hardest text-based reasoning, coding, and long-running agent tasks. Choose Plus when image or video input is required, or when a balanced capability-cost profile is more important than the highest Qwen reasoning tier.
Why does the current model accept text only when another Qwen3.7-Max snapshot supports vision?
The moving qwen3.7-max alias currently corresponds to the text-only qwen3.7-max-2026-05-20 snapshot. Alibaba separately lists qwen3.7-max-2026-06-08 as a multimodal snapshot; capabilities should not be transferred between these IDs without confirming which one an endpoint serves.
When should a production application pin a snapshot instead of using qwen3.7-max?
Pin a dated snapshot when regression testing, reproducibility, or a fixed modality contract matters. Use the moving alias when receiving future model updates is desirable and the application has evaluation checks that can detect behavioral changes.
How should I evaluate Qwen3.7-Max before assigning it long autonomous tasks?
Test it on representative end-to-end jobs, including tool failures, ambiguous instructions, context growth, and recovery after an incorrect step. Measure completion quality and verification behavior, not only the quality of the first response.