DeepSeek-V3.2
DeepSeek- Arena score
- 1425
- Among open models
- #29of 148
- Among all on Arena
- #108
- Votes
- 47,778
Measured on DeepSeek-V3.2, the served version, not a local build. LMArena, CC BY 4.0, published Nov 11, 2026.
Score by category
CategoryScoreOpen rankArena rankVotes
- Chinese1455#38#1213.7K
- Coding1453#29#1018.8K
- Software & IT1450#30#10614.2K
- Hard prompts, English1446#25#979.5K
- Science1446#25#917.5K
- Medicine & health1441#30#993.1K
- English1440#27#10017K
- Applied math1439#29#941.9K
- Expert1435#32#1123.4K
- Math1435#25#923.1K
- Hard prompts1434#28#10626.9K
- French1429#36#1201.2K
- Long prompts1428#23#9513.6K
- Multi-turn1427#27#1108.3K
- Legal & government1425#34#1243.5K
- Spanish1424#33#1091.5K
- Russian1424#21#956.1K
- Polish1419#28#1051.1K
- Excluding ties1415#28#10633.8K
- Instructions1413#25#10113.5K
- German1412#30#113741
- Business & finance1410#40#1259K
- Non-English1409#29#11128.3K
- Writing & literature1407#23#10210.6K
- Creative writing1400#24#1046.8K
- Media & sports1395#28#1118.7K
- Japanese1377#28#106382
- Korean1370#28#114791
About as strong, smaller
Compare
gemma-4-26B-A4B-itGoogle143425.8B

Qwen3.8-27BQwen143727.8B

gemma-4-31B-itGoogle144331.3B

Qwen3-Next-80B-A3B-InstructQwen141781.3B
Discussion
Built on it
Ways to run it
- GGUF · llama.cpp, Ollama, LM Studiounsloth/DeepSeek-V3.2-GGUF6 builds
- MLX · Apple silicon: mlx-lm, LM Studiobaa-ai/DeepSeek-V3.2-RAM-350GB-MLX2 builds
- AWQ · vLLM, SGLangQuantTrio/DeepSeek-V3.2-AWQ1 build
Most downloaded
BenchClub runs
Not runnable yet
Runs open once a DeepSeek-V3.2 build is registered.
More from DeepSeek
See more
DeepSeek-V4-Flash-07312M304B

DeepSeek-V4.1-Flash1.6M763B

DeepSeek-V4-Flash1.2M291B

DeepSeek-V4-Pro309.2K1.6T
Other >300B models
See more
GLM-5.3-FlashZ.ai6.4M321B
InklingThinking Machines1.5M952B
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16NVIDIA1M561B

Kimi-K3Moonshot AI768.9K2.8T