Step3 Vl 10B step3-vl-10b

stepfun-ai · stepfun-ai

#81 overall

params
10.2B
arch
dense
license
apache-2.0 open weights
released
Jan 2026
downloads/30d
9.7k

# architecture

attn shared GQA 32q/8kv feed-forward (dense) every parameter, every token out ×36 layers 10.2B — no routing
layers
36
d_model
4096
heads
32q / 8kv
head dim
128
vocab
151936
family
qwen3

signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.

# why ranked

Overall: #81 — score 28.0 ● high — all signals present

signalweightinput (0–100)
aa_intelligence 0.70 14.6
bench_composite 0.30 59.0

benchmark panel evidence:

complete panelscore (0–100)
aime-2025-v159.0

full methodology

# trend

OGM score · last 41 days
OGM score over 41 days
overall rank (up is better)
Overall rank over 41 days

# benchmarks

benchmarkscoresourcedate
AA Intelligence Index 7.7 Artificial Analysis —
AIME 2025 0.9 / 1 LLM Stats —

# where to run

no known API providers — download the weights from Hugging Face and run them yourself.

source aliases
llmstats
step3-vl-10b