DeepSeek V3.2 deepseek-v3.2
deepseek-ai · deepseek hybrid-reasoning flagship
#10 overall #6 coding #9 agentic
- params
- 671B (A37B)
- arch
- moe
- context
- 160k
- license
- mit open weights
- released
- Dec 2025
- reasoning
- yes
- downloads/30d
- 1.9M
# why ranked
Overall: #10 — score 43.1 ● medium — one signal missing, weights renormalized
| signal | weight | input (0–100) |
|---|---|---|
| aa_intelligence | 0.50 | 33.4 |
| arena_elo | 0.30 | — |
| bench_composite | 0.20 | 67.5 |
missing signals are dropped and the remaining weights renormalized — never imputed.
Coding: #6 — score 62.3 ● high — all signals present
| signal | weight | input (0–100) |
|---|---|---|
| aa_coding | 0.25 | 51.7 |
| aider_polyglot | 0.25 | 100.0 |
| arena_elo_code | 0.15 | — |
| swe_bench_verified | 0.35 | 42.9 |
missing signals are dropped and the remaining weights renormalized — never imputed.
Agentic: #9 — score 42.9 ● low — a single signal; treat with caution
| signal | weight | input (0–100) |
|---|---|---|
| swe_bench_verified | 1.00 | 42.9 |
# trend
collecting history — trends appear after a few daily runs.
# benchmarks
| benchmark | score | source | date |
|---|---|---|---|
| AA Coding Index | 43.7 | Artificial Analysis | — |
| AA Intelligence Index | 33.4 | Artificial Analysis | — |
| AA Math Index | 92.0 | Artificial Analysis | — |
| Aider Polyglot | 74.2 | Aider | Oct 2025 |
| Aider-polyglot | 0.7 / 1 | LLM Stats | — |
| AIME 2025 | 0.9 / 1 | LLM Stats | — |
| Browsecomp | 0.5 / 1 | LLM Stats | — |
| Browsecomp-zh | 0.7 / 1 | LLM Stats | — |
| Codeforces | 0.8 / 3000 | LLM Stats | — |
| Simpleqa | 1.0 / 1 | LLM Stats | — |
| SWE-bench Verified | 60.0 | SWE-bench | Dec 2025 |
# where to run
| provider | quant | ctx | $/M in | $/M out | $/M blend | tps | uptime |
|---|---|---|---|---|---|---|---|
| StreamLake | fp8 | 125k | $0.23 | $0.34 | $0.26 | — | 100.0% |
| Baidu | fp8 | 128k | $0.25 | $0.38 | $0.28 | — | 100.0% |
| AtlasCloud | fp8 | 160k | $0.26 | $0.38 | $0.29 | — | 99.9% |
| DeepInfra | fp4 | 160k | $0.26 | $0.38 | $0.29 | — | 98.6% |
| SiliconFlow | fp8 | 160k | $0.26 | $0.42 | $0.30 | — | 99.8% |
| Novita | fp8 | 160k | $0.27 | $0.40 | $0.30 | — | 100.0% |
| AtlasCloud | fp8 | 160k | $0.27 | $0.41 | $0.30 | — | 100.0% |
| Novita | fp8 | 160k | $0.27 | $0.41 | $0.30 | — | 100.0% |
| SiliconFlow | fp8 | 160k | $0.27 | $0.41 | $0.30 | — | 100.0% |
| GMICloud | fp8 | 160k | $0.29 | $0.43 | $0.32 | — | 99.5% |
| Venice | unknown | 160k | $0.33 | $0.48 | $0.37 | — | 98.6% |
| Alibaba | unknown | 128k | $0.37 | $1.11 | $0.56 | — | 99.3% |
| DigitalOcean | unknown | 64k | $0.42 | $1.36 | $0.66 | — | 95.3% |
| Friendli | unknown | 160k | $0.50 | $1.50 | $0.75 | — | 99.9% |
| unknown | 160k | $0.56 | $1.68 | $0.84 | — | 99.7% | |
| Phala | unknown | 160k | $1.00 | $1.00 | $1.00 | — | 100.0% |
| SambaNova | unknown | 32k | $3.00 | $4.50 | $3.38 | — | — |
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)- llmstats
deepseek-v3.2-exp- openrouter
deepseek/deepseek-chat-v3.2,deepseek/deepseek-v3.2,deepseek/deepseek-v3.2-exp- swebench
deepseek-v3.2-reasoner