DeepSeek V3.1 deepseek-v3.1

deepseek-ai · deepseek hybrid-reasoning flagship

#30 overall

params
671B (A37B)
arch
moe
context
160k
license
mit open weights
released
Aug 2025
reasoning
yes
downloads/30d
279.2k

# why ranked

Overall: #30 — score 21.9 ● medium — one signal missing, weights renormalized

signalweightinput (0–100)
aa_intelligence 0.50 21.0
arena_elo 0.30
bench_composite 0.20 24.2

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

full methodology

# trend

collecting history — trends appear after a few daily runs.

# benchmarks

benchmarkscoresourcedate
AA Intelligence Index 21.0 Artificial Analysis
AA Math Index 89.7 Artificial Analysis
Aider-polyglot 0.7 / 1 LLM Stats
AIME 2024 0.7 / 1 LLM Stats
AIME 2025 0.5 / 1 LLM Stats
Browsecomp 0.3 / 1 LLM Stats
Browsecomp-zh 0.5 / 1 LLM Stats
Codeforces 0.7 / 3000 LLM Stats

# where to run

providerquantctx$/M in$/M out$/M blendtpsuptime
DeepInfrafp4 160k $0.21$0.79$0.35 98.2%
Novitafp8 128k $0.27$1.00$0.45 100.0%
SiliconFlowfp8 160k $0.27$1.00$0.45 99.0%
AtlasCloudfp8 128k $0.30$0.95$0.46 100.0%
WandBfp8 161k $0.55$1.65$0.82 100.0%
SambaNovafp8 128k $0.65$1.50$0.86 100.0%
Googleunknown 160k $0.60$1.70$0.88 100.0%
Maraunknown 128k $0.60$1.70$0.88 72.5%

blended = (3·input + output) / 4 per 1M tokens · cheapest first

# variants

DeepSeek V3.1 Terminus

source aliases
aa
deepseek-v3.1
arena
DeepSeek-V3.1
openrouter
deepseek/deepseek-chat-v3.1