What are the hardware requirements for Qwen3 235B A22B Thinking 2507?

Qwen3 235B A22B Thinking 2507 requires approximately 167.9 GB of VRAM and supports a context window of 256K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Qwen3 235B A22B Thinking 2507 and how large is it?

Qwen3 235B A22B Thinking 2507 is developed by unsloth, a model with 235b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does Qwen3 235B A22B Thinking 2507 perform on standard benchmarks?

Qwen3 235B A22B Thinking 2507 has the following published scores: LMSYS ELO 1399. Compare against reference models on this page or on the LLM Explorer leaderboards.

Qwen3 235B A22B Thinking 2507 by unsloth — VRAM 167.9GB, 256K context

Name: Qwen3 235B A22B Thinking 2507
Author: unsloth

Qwen3 235B A22B Thinking 2507 is an open-source language model by unsloth. Features: 235b LLM, VRAM: 167.9GB, Context: 256K, License: apache-2.0, LLM Explorer Score: 0.33, ELO: 1399.

Arxiv:2505.09388 Base model:finetune:qwen/qwen3... Base model:qwen/qwen3-235b-a22... Conversational Endpoints compatible Qwen3 moe Region:us Safetensors Sharded Tensorflow Unsloth

Model Card on HF 🤗: https://huggingface.co/unsloth/Qwen3-235B-A22B-Thinking-2507

Qwen3 235B A22B Thinking 2507 Benchmarks

LMSys ELO 1399 vs 1272 (so35) 10%

^nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").

What is the LLM Explorer Rank (Score)

Qwen3 235B A22B Thinking 2507 Parameters and Internals

LLM Name	Qwen3 235B A22B Thinking 2507
Repository 🤗	https://huggingface.co/unsloth/Qwen3-235B-A22B-Thinking-2507
Base Model(s)	Qwen3 235B A22B Thinking 2507 Qwen/Qwen3-235B-A22B-Thinking-2507
Model Size	235b
Required VRAM	167.9 GB
Updated	2026-07-13
Maintainer	unsloth
Model Type	qwen3_moe
Model Files	4.0 GB: 1-of-118 4.0 GB: 2-of-118 4.0 GB: 3-of-118 4.0 GB: 4-of-118 4.0 GB: 5-of-118 4.0 GB: 6-of-118 4.0 GB: 7-of-118 4.0 GB: 8-of-118 4.0 GB: 9-of-118 4.0 GB: 10-of-118 4.0 GB: 11-of-118 4.0 GB: 12-of-118 4.0 GB: 13-of-118 4.0 GB: 14-of-118 4.0 GB: 15-of-118 4.0 GB: 16-of-118 4.0 GB: 17-of-118 4.0 GB: 18-of-118 4.0 GB: 19-of-118 4.0 GB: 20-of-118 4.0 GB: 21-of-118 4.0 GB: 22-of-118 4.0 GB: 23-of-118 3.9 GB: 24-of-118 4.0 GB: 25-of-118 4.0 GB: 26-of-118 4.0 GB: 27-of-118 4.0 GB: 28-of-118 4.0 GB: 29-of-118 4.0 GB: 30-of-118 4.0 GB: 31-of-118 4.0 GB: 32-of-118 4.0 GB: 33-of-118 4.0 GB: 34-of-118 4.0 GB: 35-of-118 4.0 GB: 36-of-118 4.0 GB: 37-of-118 4.0 GB: 38-of-118 4.0 GB: 39-of-118 4.0 GB: 40-of-118 4.0 GB: 41-of-118 4.0 GB: 42-of-118
Model Architecture	Qwen3MoeForCausalLM
License	apache-2.0
Context Length	262144
Model Max Length	262144
Transformers Version	4.54.0
Vocabulary Size	151936
Torch Data Type	bfloat16

Rank the Qwen3 235B A22B Thinking 2507 capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Qwen3 235B A22B Thinking 2507

Best Alternatives	Context / RAM	Downloads	Likes
...en3 235B A22B Instruct 2507 1M	986K / 171.9 GB	4	1
Qwen3 235B A22B Instruct 2507	256K / 167.9 GB	28	10
Qwen3 235B A22B Instruct 2507	256K / 171.9 GB	59472	793
Vivo C V1	256K / 156.4 GB	415	2
... 235B A22B Thinking 2507 NVFP4	256K / 139.2 GB	22828	8
Qwen3 235B A22B Thinking 2507	256K / 171.9 GB	20387	408
... 235B A22B Instruct 2507 NVFP4	256K / 139.2 GB	14322	10
...n3 235B A22B Instruct 2507 FP8	256K / 236.5 GB	54126	147
MiroThinker 1.7 Mini	256K / 61.1 GB	584	104
...n3 235B A22B Thinking 2507 FP8	256K / 236.5 GB	46813	86

Note: green Score (e.g. "73.2") means that the model is better than unsloth/Qwen3-235B-A22B-Thinking-2507.