Qwen AgentWorld 35B A3B
Qwen/Qwen-AgentWorld-35B-A3B
Open Source · chat · open-weights · seen 14m ago
Context
262.1K
Max output
—
Input $/1M
$0.14
Output $/1M
$1.00
Modalities
text
Released
22 Jun 2026
License: apache-2.0 · Qwen/Qwen-AgentWorld-35B-A3B
AI summary
● machine-written
Qwen AgentWorld 35B A3B released as open-weight agent model
Qwen AgentWorld 35B A3B is an open-weight multimodal model with 35 billion total parameters and 3 billion active parameters per token, designed for agent-based applications with extended context. The model uses a hybrid sparse mixture-of-experts architecture with Gated DeltaNet linear attention and supports text, image, and video inputs. It includes integrated thinking mode with reasoning traces and function calling capabilities, released under Apache 2.0 license.
What's new
- 35B total parameters with 3B active per token via sparse MoE architecture
- 262K token native context window, extensible to 1M via YaRN
- Integrated thinking mode with reasoning traces across multi-turn conversations
- Supports text, image, and video modalities
- Released April 27, 2026 under Apache 2.0 license
Best for
Multi-turn agent-based applications requiring extended contextTasks leveraging vision and text inputsFunction calling and structured output generationEfficient inference at reduced compute cost