DeepSeek V4 Pro deepseek-v4-pro

deepseek-ai · deepseek reasoning flagship

#2 overall

params
862B
arch
moe
context
1024k
license
mit open weights
released
Apr 2026
reasoning
yes
train compute
9.7e+24 FLOP
downloads/30d
652.3k

# architecture

attn shared · reasoning GQA 128q/1kv router top-6 of 384 +1 shared expert ×384 expert out ×61 layers ctx 1,048,576 862B pool
layers
61
d_model
7168
heads
128q / 1kv
head dim
512
experts
top-6 of 384 + 1 shared
vocab
129280
family
deepseek_v4

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: #2 — score 96.1 ● high — all signals present

signalweightinput (0–100)
aa_intelligence 0.70 95.9
bench_composite 0.30 96.7

benchmark panel evidence:

complete panelscore (0–100)
livebench-v196.7

Coding: provisional — score 92.1 ● 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_coding 0.30 92.1
aider_polyglot 0.30
swe_bench_verified 0.40

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

full methodology

# trend

OGM score · last 55 days
OGM score over 55 days
overall rank (up is better)
Overall rank over 55 days

methodology changed during this history window; score movement across that boundary is not model movement. See methodology v5.

# benchmarks

benchmarkscoresourcedate
AA Coding Index 68.8 Artificial Analysis
AA Intelligence Index 42.1 Artificial Analysis
LiveBench Agentic Coding 42.6 LiveBench Jun 2026
LiveBench Coding 70.0 LiveBench Jun 2026
LiveBench Data Analysis 74.5 LiveBench Jun 2026
LiveBench IF 62.4 LiveBench Jun 2026
LiveBench Language 78.1 LiveBench Jun 2026
LiveBench Mathematics 90.7 LiveBench Jun 2026
LiveBench Reasoning 82.7 LiveBench Jun 2026
LMArena Elo 1457.0 LMArena Aug 2026
LMArena Elo (Code) 1445.0 LMArena Aug 2026

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime
Nebius nebius
DeepSeek 1000k $0.43$0.87$0.00 via modelsdev
Nvidia 1024k $0.43$0.87$0.00 via modelsdev
Tencent 1000k $0.43$0.87 via litellm
DigitalOceanunknown 1024k $0.87$1.74$0.17 via openrouter 97.8%
Baidufp8 1024k $0.90$1.80$0.07 via openrouter 99.9%
StreamLakefp8 1000k $0.90$1.80$0.08 via openrouter 98.7%
Fireworksunknown 1024k $1.20$1.20$0.60 via openrouter
GMICloudfp8 1024k $1.04$2.09$0.09 via openrouter 98.5%
Ionstreamfp4 1024k $1.13$2.26$0.09 via openrouter 99.9%
Wandb 1024k $1.15$2.55 via litellm
DeepInfrafp8 1024k $1.30$2.60 deepinfra 99.1%
Alibabafp8 1000k $1.42$2.83$0.12 via openrouter 100.0%
SiliconFlowfp8 1024k $1.50$3.13$0.14 openrouter 99.2%
Novitafp8 1024k $1.60$3.20 novita 100.0%
Veniceunknown 1000k $1.65$3.30 venice 98.6%
AtlasCloudfp4 1024k $1.68$3.38 atlascloud 99.8%
BaseTenfp4 1024k $1.74$3.48$0.14 via openrouter 99.4%
BaseTenfp4 1024k $1.74$3.48$0.14 via openrouter 99.2%
NextBitfp8 1024k $1.74$3.48$0.15 via openrouter 99.6%
Parasailfp8 1024k $1.74$3.48$0.10 via openrouter 98.5%
Together 500k $1.74$3.48$0.20 via modelsdev
Azureunknown 1024k $1.91$3.83$0.16 via openrouter 99.1%
Qwen Ai Platform 1000k $2.40$4.80 via litellm
Qwencloud 1000k $2.40$4.80 via litellm

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
aa
deepseek-v4-pro
epoch
DeepSeek-V4-Pro
hf
deepseek-ai/DeepSeek-V4-Pro-DSpark
openrouter
deepseek/deepseek-v4-pro