{"id":"clover-35b-a3b-preview","slug":"clover-35b-a3b-preview","name":"Clover-35B-A3B-Preview","orgId":"diffuspace","family":"clover-35b-a3b-preview","releaseDate":null,"availability":"announced","openWeights":true,"license":null,"paramsTotal":35000000000,"paramsSource":"slug","contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://x.com/DiffuSpace_Labs/status/2115288124691820738","hfRepo":null,"modalities":["text"],"aliases":["Clover-35B-A3B-Preview","clover-35b-a3b-preview","clover 35b a3b preview","clover-35b-a3b","clover 35b","diffuspace/clover-35b-a3b-preview","diffuspace clover"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-10-28","announced":"2026-10-28","blurb":"DiffuSpace's first model: a text-only diffusion language model that writes in blocks, filling masked positions in parallel instead of one token at a time. It is derived from Qwen3.6-35B-A3B (35B total, about 3B active, mixture of experts), supports a thinking mode and 128K context, and is published on Hugging Face and ModelScope under Apache-2.0.","api":"none","openWeights":"yes","params":{"text":"35B total, about 3B active (34.66B in the safetensors)","source":"DiffuSpace's Hugging Face model card and safetensors metadata"},"facts":[{"label":"Lab","value":"DiffuSpace (team behind HKU NLP's Dream 7B diffusion model, per the launch blog)"},{"label":"Type","value":"Block-diffusion language model (dLLM), text only, with a thinking mode"},{"label":"Status","value":"Preview"},{"label":"Base model","value":"Qwen/Qwen3.6-35B-A3B"},{"label":"Architecture","value":"Hybrid Gated DeltaNet / full-attention MoE, 40 layers, 256 routed experts, top-8 routing plus 1 shared expert; bidirectional attention within each block"},{"label":"Context","value":"131,072 tokens (prompt and response together)"},{"label":"Weights","value":"Published on Hugging Face and ModelScope under Apache-2.0; BF16 weights are about 69 GB"},{"label":"Serving","value":"SGLang and llama.cpp implementations in the GitHub repo DiffuSpace/Clover-35B-A3B-Preview, plus a Transformers path"},{"label":"Hosted access","value":"Online access by application (chat.diffuspace.ai); no public API announced"}],"claims":["Lab claim: 680 tokens/s overall on internal agent workloads on one NVIDIA GB300 (BF16, batch size 1), and 515 tokens/s on agent tasks on one H200, against 365 tokens/s for a DFlash2-accelerated Qwen setup.","Lab claim: 73.6 on SWE-bench Verified, 51.7 on Terminal-Bench 2.1, 65.3 on τ³ Overall, 83.8 on GPQA Diamond and 84.8 on MMLU-Pro, ahead of LLaDA2.2 flash, Mercury 2.5, Qwen3.6-35B-A3B and Qwen3.5-35B-A3B on its agent benchmarks.","Lab claim (DiffuSpace on X): the first open-weight production diffusion language model (dLLM)."],"sources":[{"kind":"lab","title":"DiffuSpace announcement on X","url":"https://x.com/DiffuSpace_Labs/status/2115288124691820738","date":"2026-10-28"},{"kind":"lab","title":"DiffuSpace blog: Clover-35B-A3B-Preview","url":"https://diffuspace.ai/blog/clover-35b/"},{"kind":"repo","title":"Clover-35B-A3B-Preview on Hugging Face","url":"https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview","date":"2026-10-28"},{"kind":"repo","title":"Clover-35B-A3B-Preview on ModelScope","url":"https://modelscope.cn/models/Diffuspace/Clover-35B-A3B-Preview","date":"2026-10-28"},{"kind":"repo","title":"DiffuSpace/Clover-35B-A3B-Preview on GitHub (code and inference backends)","url":"https://github.com/DiffuSpace/Clover-35B-A3B-Preview"}],"peer":{"slug":"qwen3-6-35b-a3b","label":"its base model, Qwen3.6 35B A3B"},"firstBenchmark":null},"organization":{"id":"diffuspace","name":"DiffuSpace","slug":"diffuspace","country":null,"url":"https://diffuspace.ai","color":"#11110F","mark":"D"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 28, 2026; no independent results yet.","lead":"As of Nov 3, 2026, Clover-35B-A3B-Preview is pending benchmarks on isitsota: announced Oct 28, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}