Nemotron 3 Nano 30B A3B nemotron-3-nano-30b-a3b

nvidia · nvidia

unranked — no quality signals yet

params
31.6B
arch
moe
context
256k
license
other open weights
released
Dec 2025
train compute
4.8e+23 FLOP
downloads/30d
858.6k

# architecture

attn shared GQA 32q/2kv router top-6 of 128 +1 shared expert ×128 expert out ×52 layers ctx 262,144 31.6B pool
layers
52
d_model
2688
heads
32q / 2kv
head dim
128
experts
top-6 of 128 + 1 shared
vocab
131072
family
nemotron_h

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: provisional — score 94.3 ● low — a single signal; treat with caution

provisional: too few independent signals for a numbered position — the score is shown, but this model sorts after every ranked model.

signalweightinput (0–100)
aa_intelligence 0.70
bench_composite 0.30 94.3

benchmark panel evidence:

complete panelscore (0–100)
aime-2025-v194.3

missing signals are dropped and the remaining weights renormalized — never imputed.

full methodology

# trend

OGM score · last 47 days
OGM score over 47 days

# benchmarks

benchmarkscoresourcedate
AIME 2025 1.0 / 1 LLM Stats
Arena-hard-v2 0.7 / 1 LLM Stats

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime
Nvidia 256k via requesty
Crusoefp8 256k $0.05$0.20$0.03 via openrouter 100.0%
DeepInfrafp4 256k $0.05$0.20 deepinfra 99.8%
Novitafp4 256k $0.05$0.20 novita 99.9%
Nebiusfp8 256k $0.06$0.24 via openrouter 98.9%
Venicefp8 125k $0.07$0.30 venice

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
epoch
Nemotron 3-Nano-30B-A3B
llmstats
nemotron-3-nano-30b-a3b
openrouter
nvidia/nemotron-3-nano-30b-a3b