Step 3.5 Flash step-3.5-flash

stepfun-ai · stepfun-ai

#33 overall

params
199.4B
arch
moe
context
256k
license
apache-2.0 open weights
released
Feb 2026
downloads/30d
259.9k

# architecture

attn shared MHA ×64 router top-8 of 288 expert ×288 expert out ×45 layers ctx 262,144 199.4B pool
layers
45
d_model
4096
heads
64q / 64kv
head dim
128
experts
top-8 of 288
vocab
128896
family
step3p5

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: #33 — score 64.8 ● high — all signals present

signalweightinput (0–100)
aa_intelligence 0.70 57.3
bench_composite 0.30 82.3

benchmark panel evidence:

complete panelscore (0–100)
aime-2025-v184.0
browsecomp-v180.5

full methodology

# trend

OGM score · last 42 days
OGM score over 42 days
overall rank (up is better)
Overall rank over 42 days

# benchmarks

benchmarkscoresourcedate
AA Intelligence Index 17.0 Artificial Analysis —
AIME 2025 1.0 / 1 LLM Stats —
Browsecomp 0.7 / 1 LLM Stats —

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime✓
Featherless — ——— via hfrouter — —
DeepInfra 256k $0.10$0.30— via hfrouter 13 —
SiliconFlowfp8 256k $0.10$0.30— openrouter — 100.0% ✓

sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price

source aliases
llmstats
step-3.5-flash
openrouter
stepfun/step-3.5-flash