HOT RADAR
llmHEAT 10.4°

AI CURATED HOT STORY · 11/19/2026

DiffuSpace Open-Sources Clover 35B Diffusion Language Model for Agents

1 reports archived1 independent sourcesupdated 11/19/2026, 16:59:00
Synthesis & Latest Updates
1 Sources Cross-Validated

DiffuSpace has open-sourced Clover-35B-A3B-Preview, a 35B-parameter diffusion language model with roughly 3B active parameters that uses block-wise diffusion for parallel token generation, reaching 5-10x the throughput of comparable autoregressive models. Under a unified evaluation protocol it tops all seven agent benchmarks reported, scoring 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1, with the technical report and SFT code released alongside. The team expects diffusion LMs to replace same-scale autoregressive models in some scenarios within one to two years.

LATEST/DiffuSpace has open-sourced Clover-35B-A3B-Preview, a 35B-parameter diffusion language model with roughly 3B active parameters that uses block-wise diffusion for parallel token generation, reaching 5-10x the throughput of comparable autoregressive models. Under a unified evaluation protocol it tops all seven agent benchmarks reported, scoring 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1, with the technical report and SFT code released alongside. The team expects diffusion LMs to replace same-scale autoregressive models in some scenarios within one to two years.

Hourly heat curve

7 fully observed hours
Now
10.4
Peak
12.017:00
24h change
–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. InfoQ 中文T2·78 pts
    • Clover has 35B total and ~3B active parameters, using block-wise diffusion to cut redundant compute and enable bidirectional interaction within each generated block
    • It leads all seven reported agent benchmarks with 73.6 on SWE-bench Verified and 51.7 on Terminal-Bench 2.1, and generates 5-10x faster than comparable autoregressive models in some configurations
    • The team will keep most resources on foundation model training through 2027, prioritizing scaling and multimodality, while converting generation speed into task-level efficiency remains unproven