qwen3-30b-a3bQwen3 is the latest generation in the Qwen large language model series, featuring both dense and Mixture-of-Experts (MoE) architectures designed to excel in reasoning, multilingual understanding, and advanced agentic tasks. A defining capability of Qwen3 is its ability to seamlessly switch between a thinking mode for complex, multi-step reasoning and a non-thinking mode for efficient, high-quality dialogue—delivering strong versatility across use cases. Compared with earlier models such as QwQ and Qwen2.5, Qwen3 demonstrates substantial performance gains in mathematics, coding, commonsense reasoning, creative writing, and interactive conversation. The Qwen3-30B-A3B variant comprises 30.5B total parameters with 3.3B activated, 48 layers, and 128 experts (with 8 activated per task). With support for up to 131K token context lengths via YaRN, it sets a new benchmark for open-source MoE models in both capability and efficiency.
Select an endpoint and copy a working example for this model.
from openai import OpenAI client = OpenAI( api_key="YOUR_API_KEY", base_url="https://api.apertis.ai/v1") response = client.chat.completions.create( model="qwen3-30b-a3b", messages=[ {"role": "user", "content": "Hello!"} ], max_tokens=1024, temperature=0.7) print(response.choices[0].message.content) # Optional: Enable context compression to reduce token usage# response = client.chat.completions.create(# model="qwen3-30b-a3b",# messages=[{"role": "user", "content": "Hello!"}],# extra_body={"compression": {"enabled": True, "model": "gpt-4.1-mini"}}# )modelmessagesmax_tokenstemperaturetop_pstreamtoolsreasoning_effortstream_optionsthinkingextra_bodyUse these namespaced identifiers in Cursor IDE to avoid conflicts with built-in models.
See how this model compares to others from the same provider.
Qwen3.7-Plus is a cost-effective multimodal model in Alibaba's Qwen3.7 series, supporting text and image inputs with text output. It combines the series' strong language capabilities with significantly enhanced vision-language understanding, while retaining full-stack agent-level intelligence for coding, tool use, and productivity workflows. Its standout capability is multimodal interactive agency—the ability to perceive real-world scenes, understand screens and graphical interfaces, generate code from visual references, and perform end-to-end navigation within applications. This makes Qwen3.7-Plus well suited for GUI automation, visual coding, productivity agents, and multimodal task execution.
Qwen3.7-Max is the flagship model in Alibaba's Qwen3.7 series, designed for agent-centric workloads with strong performance in coding, productivity, and long-horizon autonomous execution. It supports text input and output and delivers notable improvements in coding and agentic capabilities over previous Qwen generations. Optimized for real-world workflows, the model also supports explicit prompt caching for efficient reuse of repeated context, making it well suited for scalable development, office automation, and advanced agent systems.
Qwen3.6-Max-Preview is a proprietary frontier model from Alibaba Cloud built on a sparse Mixture-of-Experts (MoE) architecture with approximately 1 trillion parameters. It is optimized for agentic coding, tool use, and long-context reasoning, supporting a 262K token context window. The model includes an integrated thinking mode that preserves reasoning across multi-turn interactions, along with support for structured outputs and function calling. Available exclusively via Alibaba Cloud Model Studio and Qwen Studio APIs, it is designed for high-performance, production-grade agent workflows.
See how this model compares to others from the same provider.
Qwen3-Coder-Next is an open-weight causal language model purpose-built for coding agents and local development workflows. It employs a sparse Mixture-of-Experts (MoE) architecture with 80B total parameters and only 3B activated per token, achieving performance comparable to models with 10–20× higher active compute. This efficiency makes it especially well suited for cost-sensitive, always-on agent deployments. Trained with a strong agentic focus, Qwen3-Coder-Next performs reliably on long-horizon coding tasks, complex tool interactions, and robust recovery from execution failures. With a native 256K context window, it integrates smoothly into real-world CLI and IDE environments and aligns well with common agent scaffolding used by modern coding tools. The model operates exclusively in non-thinking mode and does not emit <think> blocks, simplifying production integration for coding agents.
Qwen Plus 0728 is a hybrid reasoning model built on the Qwen3 foundation, offering a 1M-token context window with a balanced mix of performance, speed, and cost efficiency.
Qwen3-Next-80B-A3B-Thinking is a reasoning-first chat model designed for difficult multi-step tasks such as math proofs, code synthesis and debugging, logical reasoning, and agentic planning. It outputs structured thinking traces by default, emphasizing stability over long chains of thought, efficient inference scaling, and strong instruction following. Suited for agent frameworks, tool use, retrieval-heavy workflows, and benchmarks requiring step-by-step solutions, it supports long detailed outputs and faster generation via throughput-optimized techniques, operating exclusively in thinking mode.
Qwen3.5-397B-A17B is a native vision-language model built on a hybrid architecture that combines linear attention mechanisms with a sparse Mixture-of-Experts (MoE) design to achieve higher inference efficiency at large scale. It delivers state-of-the-art performance across a broad range of tasks, including language understanding, logical reasoning, code generation, agent-based workflows, image and video understanding, and GUI interaction. With strong coding and agent capabilities, Qwen3.5-397B-A17B demonstrates robust generalization across diverse multimodal and agentic scenarios, making it well suited for advanced applications that require integrated reasoning across text, vision, and interactive environments.
Initialized observational baseline with no recorded failures