What are the hardware requirements for 60B MoE Coder V3?

60B MoE Coder V3 requires approximately 121.8 GB of VRAM and supports a context window of 195K tokens. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed 60B MoE Coder V3 and how large is it?

60B MoE Coder V3 is developed by cloudyu, a model with 60.8b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

How does 60B MoE Coder V3 perform on standard benchmarks?

60B MoE Coder V3 has the following published scores: MMLU 75.37. Compare against reference models on this page or on the LLM Explorer leaderboards.

60B MoE Coder V3 by cloudyu — VRAM 121.8GB, 195K context

Name: 60B MoE Coder V3
Rating: 4 (1 reviews)
Author: cloudyu

60B MoE Coder V3 is an open-source language model by cloudyu. Features: 60.8b LLM, VRAM: 121.8GB, Context: 195K, License: other, MoE, Quantized, Code Generating, LLM Explorer Score: 0.12, Arc: 71.2, HellaSwag: 85.4, MMLU: 75.4, GSM8K: 66.9.

Codegen Conversational Endpoints compatible Fp16 Mixtral Moe Quantized Region:us Safetensors Sharded Tensorflow Yi

Model Card on HF 🤗: https://huggingface.co/cloudyu/60B_MoE_Coder_v3

60B MoE Coder V3 Parameters and Internals

Model Type

chat model, general purpose

Use Cases

Areas:

coding, general purpose

Supported Languages

English (proficient), Chinese (proficient)

LLM Name	60B MoE Coder V3
Repository 🤗	https://huggingface.co/cloudyu/60B_MoE_Coder_v3
Model Size	60.8b
Required VRAM	121.8 GB
Updated	2026-07-18
Maintainer	cloudyu
Model Type	mixtral
Model Files	4.9 GB: 1-of-25 4.8 GB: 2-of-25 4.9 GB: 3-of-25 4.8 GB: 4-of-25 4.9 GB: 5-of-25 4.8 GB: 6-of-25 4.9 GB: 7-of-25 5.0 GB: 8-of-25 5.0 GB: 9-of-25 5.0 GB: 10-of-25 5.0 GB: 11-of-25 5.0 GB: 12-of-25 5.0 GB: 13-of-25 5.0 GB: 14-of-25 5.0 GB: 15-of-25 5.0 GB: 16-of-25 5.0 GB: 17-of-25 5.0 GB: 18-of-25 5.0 GB: 19-of-25 5.0 GB: 20-of-25 5.0 GB: 21-of-25 5.0 GB: 22-of-25 5.0 GB: 23-of-25 5.0 GB: 24-of-25 2.8 GB: 25-of-25
Quantization Type	fp16
Generates Code	Yes
Model Architecture	MixtralForCausalLM
License	other
Context Length	200000
Model Max Length	200000
Transformers Version	4.37.2
Tokenizer Class	LlamaTokenizer
Padding Token	<s>
Vocabulary Size	64000
Torch Data Type	bfloat16

Rank the 60B MoE Coder V3 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