Clover 35B-A3B Preview

Name: Clover 35B-A3B Preview
Author: DiffuSpace

DiffuSpace · Clover · released 2026-10-24 · data checked 2026-11-04

Clover 35B-A3B Preview is a 34.7B-parameter Mixture-of-Experts model (3B active per token) from DiffuSpace (China), released 2026-10-24 under the Apache-2.0 license. It doesn't have enough reported benchmarks for an index score yet. It needs about 22.4 GB of memory at 4-bit with an 8K-token context.

A diffusion language model: it writes blocks of 32 tokens in parallel instead of one token at a time, converted from Qwen3.6 35B-A3B. Results are only published as an image (agentic benchmarks), so none are listed yet. Needs custom code; DiffuSpace recommends its SGLang build.

Small LM Index

—

Needs 2 core benchmarks (has 0)

Parameters

34.7B

3B active · MoE

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

131K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best medium model on that benchmark.

No benchmark results for this release yet. Know of one? Send it.

Specifications

Lab: DiffuSpace (CN · China)
Based on: Qwen3.6 35B-A3B (fine-tune or distillation)
Architecture: Mixture-of-Experts
Total parameters: 34.7B
Active parameters: 3B
Size class: Medium
Context window: 131K tokens
Input modalities: text
Reasoning: Hybrid (thinking on/off)
License: Apache-2.0
Openness: Open weights
Memory to run: ~22.4 GB at 4-bit, 8K context (more sizes)
Hugging Face: 3.3K downloads (30 days) · 241 likes

Weights on Hugging Face ↗

Compare with

Clover 35B-A3B Preview vs Edge0 35B-A3B (preview) (34.7B, index 63.0)
Clover 35B-A3B Preview vs Qwen3.6 35B-A3B (35B, index 68.3)
Clover 35B-A3B Preview vs Qwen3.5 35B-A3B (35B, index 66.9)

Sources

https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview

Something wrong here? Report it.

Run it locally

Uses Diffuspace/Clover-35B-A3B-Preview-GGUF (Q4_K_M). Swap the tag for another size: Q4_0, Q4_K_M, Q8_0.

Ollama

ollama run hf.co/Diffuspace/Clover-35B-A3B-Preview-GGUF:Q4_K_M

llama.cpp

llama-server -hf Diffuspace/Clover-35B-A3B-Preview-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

vLLM (GPU server)

vllm serve Diffuspace/Clover-35B-A3B-Preview

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	131K context
16-bit F16	71 GB	71 GB	73 GB
8-bit Q8_0	38 GB	39 GB	41 GB
4-bit Q4_K_M	22 GB	23 GB	25 GB

Check it against your GPU, Mac or phone →