Alibaba Releases Qwen3.7 Max with 1M Token Context Window for Agent and Coding Tasks
Alibaba has released Qwen3.7 Max, the flagship model in its Qwen3.7 series, featuring a 1 million token context window. The text-only model is designed for agent-centric workloads with strengths in coding, office productivity, and long-horizon autonomous execution, and includes explicit prompt caching support.
Qwen3.7 Max at a glance
How Qwen3.7 Max ranks now
| Benchmark | Score | Rank | Source |
|---|---|---|---|
| GPQA | 92.3% | 23 of 162 | Artificial Analysis, Oct 2026 ↗ |
| SWE-bench Verified | 80.4% | 12 of 80 | llm-stats, Jul 2026 ↗ |
| Speed (tok/s) | 158 tok/s | 17 of 100 | Artificial Analysis, Nov 2026 ↗ |
| MMLU-Pro | 89.6% | 4 of 73 | llm-stats, Jul 2026 ↗ |
Rank is against every model in our database with a sourced score on that benchmark, each counted once with its best score.
Alibaba Releases Qwen3.7 Max with 1M Token Context Window for Agent and Coding Tasks
Alibaba has released Qwen3.7 Max, the flagship model in its Qwen3.7 series, featuring a 1 million token context window. The model supports text input and output only.
Key Specifications
- Context window: 1 million tokens
- Released: May 21, 2025
- Modalities: Text only (no multimodal support)
- Prompt caching: Explicit prompt caching supported
- Pricing: Not yet disclosed
Performance Focus
According to Alibaba, Qwen3.7 Max is optimized for agent-centric workloads with three primary use cases:
- Coding tasks: The model claims notable gains in coding performance over previous Qwen generations
- Office and productivity applications: Designed for document processing and workflow automation
- Long-horizon autonomous execution: Built for multi-step agent tasks that require sustained context
The company states the model offers "notable gains in coding and agentic performance" compared to prior Qwen versions, though specific benchmark scores have not been published at launch.
Technical Features
The 1 million token context window places Qwen3.7 Max among models with extended context capabilities, comparable to recent releases from other vendors. The explicit prompt caching feature is designed to optimize performance when reusing repeated context across multiple requests, reducing latency and compute costs for agent workflows.
Parameter count and training data cutoff date have not been disclosed.
What This Means
Qwen3.7 Max represents Alibaba's continued push into the agent and coding model market with a focus on extended context. The 1M token window and prompt caching position it for complex agent workflows that require maintaining state across long interactions. However, without published benchmark scores or pricing, direct performance and cost comparisons with competing models like GPT-4, Claude 3.5 Sonnet, or DeepSeek remain unclear. The agent-first design signals Alibaba's bet on autonomous AI systems as a key use case for frontier models.
Related Articles
Nex AGI releases Nex-N2.6, a 321B-parameter agentic model with 262K context at $0.12/$0.40 per 1M tokens
Nex AGI released Nex-N2.6 on Nov 10, 2026, an agentic model built on GLM-5.3-Flash with about 321B parameters and a 262K-token context window. OpenRouter lists it at $0.12 per 1M input tokens and $0.40 per 1M output tokens, labeled 20% off.
Baidu releases ERNIE 5.5 Flash Preview: 280B parameters, 14B active, 1M context, $0.28/1M input
Baidu has released ERNIE 5.5 Flash Preview, a model with 280 billion total parameters, 14 billion active per token, and a 1 million token context window. It is listed at $0.28 per 1M input tokens and $1.13 per 1M output tokens on OpenRouter via Baidu Qianfan.
MBZUAI's IFM releases PAN 2, an Apache 2.0 world model with a 29B diffusion transformer
The Institute of Foundation Models (IFM) at MBZUAI released PAN 2, an Apache 2.0 world model. It couples an LLM planner with a 29B diffusion transformer to simulate up to 20 seconds of action outcomes as video.
Fish Audio releases Live 3, a full-duplex voice model that reacts in as little as 186 ms
Fish Audio released Live 3, a full-duplex voice model for voice agents. It listens while it speaks, separates "mm-hmm" from real interruptions, and can react in as little as 186 ms. It is English only at launch.
Comments
Loading...