DeepSeek V3.2 deepseek-v3.2

deepseek-ai · deepseek hybrid-reasoning flagship

unranked — no quality signals yet

params
671B (A37B)
arch
moe
context
128k
license
mit open weights
released
Dec 2025
reasoning
yes
downloads/30d
1.9M

# why ranked

Coding: #14 — score 64.1 ● medium — one signal missing, weights renormalized

signalweightinput (0–100)
aa_coding 0.25
aider_polyglot 0.25 100.0
arena_elo_code 0.15
swe_bench_verified 0.35 38.5

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

Agentic: #9 — score 38.5 ● low — a single signal; treat with caution

signalweightinput (0–100)
swe_bench_verified 1.00 38.5

full methodology

# trend

collecting history — trends appear after a few daily runs.

# benchmarks

benchmarkscoresourcedate
Aider Polyglot 74.2 Aider Oct 2025
SWE-bench Verified 60.0 SWE-bench Dec 2025

# where to run

providerquantctx$/M in$/M out$/M blendtpsuptime
StreamLakefp8 125k $0.23$0.34$0.26 100.0%
Baidufp8 128k $0.25$0.38$0.28 100.0%
AtlasCloudfp8 160k $0.26$0.38$0.29 100.0%
DeepInfrafp4 160k $0.26$0.38$0.29 96.1%
SiliconFlowfp8 160k $0.26$0.42$0.30 99.9%
Novitafp8 160k $0.27$0.40$0.30 99.9%
GMICloudfp8 160k $0.29$0.43$0.32 97.7%
Veniceunknown 160k $0.33$0.48$0.37 99.7%
Alibabaunknown 128k $0.37$1.11$0.56 99.7%
DigitalOceanunknown 64k $0.42$1.36$0.66 100.0%
Friendliunknown 160k $0.50$1.50$0.75 99.8%
Googleunknown 160k $0.56$1.68$0.84 99.9%
Phalaunknown 160k $1.00$1.00$1.00 99.5%
SambaNovaunknown 32k $3.00$4.50$3.38 84.7%

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

source aliases
aa
deepseek-v3.2
aider
DeepSeek-V3.2-Exp (Chat), DeepSeek-V3.2-Exp (Reasoner)
openrouter
deepseek/deepseek-chat-v3.2, deepseek/deepseek-v3.2
swebench
deepseek-v3.2-reasoner