# Clover-35B-A3B-Preview: a 35B diffusion language model for agent tasks

> Original title: 刚刚，扩散模型杀进Agent主场！350亿参数，速度比Jev快一倍

- Source: QbitAI
- Published: 2026-10-29T02:18:28.000Z
- Bad Influence: https://badinfluence.news/items/et1s84hxy5cenxwz4xsj09ke5
- Original: https://www.qbitai.com/2026/10/505934.html

## Summary

Diffusion Intelligence released Clover-35B-A3B-Preview, a diffusion language model with 35B total parameters and about 3B activated per token, which the source says can handle code generation and tool use. The source reports a median 1.52 seconds on 49 WindTunnel web-agent tasks versus 3.21 seconds for the Jev plus Mercury 2.5 combination, with 98.6% success against 95.9%. It also reports 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1.

## Why it matters

The article explains how diffusion language models generate text in parallel and how Clover adapts them for coding and tool-using agent tasks, which helps readers compare them with autoregressive models.