Alibaba

Qwen AgentWorld 35B A3B

Name: Qwen AgentWorld 35B A3B
Rating: 83 (87 reviews)
Author: Alibaba

前沿

HuggingFace

103.6K下载量629点赞Jun 2026发布日期262K tokens上下文Apache 2.0许可证76 优秀质量

Qwen AgentWorld 35B A3B (34.70000076293945B parameters) requires approximately 23.0 GB of VRAM with Q4_K_M quantization. As a Mixture of Experts model with 3B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 27 GB of VRAM.

快速开始

— 复制粘贴即可本地运行

Copy-paste commands to run Qwen AgentWorld 35B A3B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "Qwen/Qwen-AgentWorld-35B-A3B" \
  --hf-file "Qwen-AgentWorld-35B-A3B-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters34.7B (3B active)

Architecturemoe (MoE)

Context262K tokens

Modalitytext

Min RAM13.5 GB

Rec. RAM21.2 GB (Q4_K_M)

LicenseApache 2.0

FamilyQwen

✓ Reasoning

About this model

Qwen-AgentWorld is the first language world model to cover seven agent interaction domains in a single model, built on Qwen3.5-35B-A3B-Base. Rather than answering prompts, it simulates agentic environments — predicting the next environment state given an agent's action and interaction history — via long chain-of-thought reasoning, with environment modeling as the training objective from continued pre-training onward.

•Seven unified domains: MCP tool calling, Search, Terminal, SWE, Android, Web and OS.
•Native world model — environment modeling from the CPT stage, not a post-hoc add-on.
•35B total / ~3B active per token, so it simulates environments on a single GPU.
•Zero-shot generalization to out-of-distribution environments, with controllable perturbations.
•262K context; hybrid linear + full attention (1-in-4 layers) keeps KV cache small.

Qwen AgentWorld 35B A3B 的最佳选择

RTX 5090 32GBS

32 GB

RTX PRO 4500 Blackwell 32GBS

32 GB

AMD Instinct MI100 32GBS

运行此模型

Qwen AgentWorld 35B A3B on RTX 5090 32GB Qwen AgentWorld 35B A3B on RTX PRO 4500 Blackwell 32GB Qwen AgentWorld 35B A3B on AMD Instinct MI100 32GB

量化选项

各量化级别的 VRAM 估算

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q1_0_G128	1.125	5.0 GB	Very Low	—
Q2_0_G128	1.71	9.3 GB	Low	—
Q2_K	2	13.5 GB	Low	—
Q3_K_S	3	17.0 GB	Low	—
NVFP4	4	19.4 GB	Medium	—
Q4_K_M	4	21.2 GB	Medium	—
Q5_K_M	5	25.0 GB	High	—
Q6_K	6	28.5 GB	High	—
Q8_0	8	37.1 GB	Very High	—
F16	16	71.1 GB	Maximum	—

硬件兼容性

全部硬件的适配估算

打开计算器

Computing compatibility...

内存详细分析

Reference: RTX 2060 6GB

Weights21.2 GB

KV Cache0.3 GB

Runtime0.9 GB

Headroom0.6 GB

常见问题

FAQ — Qwen AgentWorld 35B A3B

另请参阅

量化指南评分方法打开计算器