Gemini 4 Argon
Claude Opus 5.5
Muse Wonder
MiMo-V2.6-Pro
Qwen3.8-Max
Kimi K3
GLM-5.3
GPT-5.5
ERNIE 5.1
Grok 4.20 Beta
amazon-nova-experimental-chat
Dola Seed 2.0 Pro
NVIDIA-Nemotron-3-Ultra-550B-A55B
Gemma 4 31B
Step 5 Preview
DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is an open-weight model from DeepSeek, released Dec 1, 2025. It ranks #105 of 351 on the LMArena text leaderboard with a rating of 1425. It costs $0.28 per million input tokens and $0.42 per million output tokens, with a 160K-token context window.
- Developer
- DeepSeek
- Released
- Dec 1, 2025
- Context window
- 160K
- Price in / out, per 1M tokens
- $0.28 / $0.42
- Weights
- Open · MIT
- Inputs
- Text
- Reasoning
- Yes
- Tool use
- Yes
- Hugging Face downloads, 30 days
- 5.6M
- Hugging Face likes
- 1.6K
The rating LMArena published at each update since Oct 3, 2025. Ratings move as votes come in and new models join.
Benchmarks
| Benchmark | Score bar | Accuracy | Rank | Cost / test |
|---|---|---|---|---|
| SWE-benchCoding | 67.60% | 67 / 85 | $0.67 | |
| Vibe Code BenchCoding | 5.11% | 102 / 115 | $2.47 | |
| LiveCodeBenchCoding | 69.86% | 92 / 143 | $0.56 / $1.68 | |
| MedQAMedicine | 89.45% | 50 / 92 | $0.56 / $1.68 | |
| GPQA DiamondReasoning | 76.26% | 81 / 138 | $0.56 / $1.68 | |
| MMLU-ProReasoning | 83.06% | 76 / 138 | $0.56 / $1.68 | |
| AIMEReasoning | 64.79% | 55 / 96 | $0.56 / $1.68 |
Vals AI, as of Oct 24, 2026.
LMArena, by category
| Leaderboard | Rating | Rank | Votes |
|---|---|---|---|
| TextLMArena | 1425 | 105 / 351 | 9K |
| CodingLMArena | 1453 | 96 / 346 | 8.9K |
| MathLMArena | 1435 | 92 / 339 | 3.1K |
| Creative writingLMArena | 1403 | 92 / 349 | 1.7K |
| Instruction followingLMArena | 1413 | 102 / 351 | 13.6K |
| Hard promptsLMArena | 1434 | 104 / 351 | 27.1K |
| Multi-turnLMArena | 1427 | 108 / 349 | 8.3K |
| Long queriesLMArena | 1428 | 95 / 330 | 13.7K |
| ExpertLMArena | 1436 | 107 / 305 | 3.4K |
| Legal and governmentIndustries | 1426 | 120 / 325 | 3.5K |
| Medicine and healthcareIndustries | 1469 | 40 / 325 | 443 |
| Business and financeIndustries | 1412 | 120 / 344 | 2.2K |
| ChineseLanguages | 1461 | 107 / 328 | 1K |
| SpanishLanguages | 1441 | 81 / 249 | 355 |
| GermanLanguages | 1440 | 73 / 261 | 227 |
| FrenchLanguages | 1434 | 112 / 252 | 227 |
| JapaneseLanguages | 1374 | 109 / 236 | 384 |
| WebDevLMArena | 1363 | 113 / 124 | 6.9K |
LMArena, CC BY 4.0, as of Oct 22, 2026. Where a model is rated at several effort levels, its best is shown (DeepSeek-V3.2-Exp (Thinking), DeepSeek-V3.2, DeepSeek-V3.2-Exp, DeepSeek-V3.2 (Thinking)).
Compare DeepSeek-V3.2-Exp
DeepSeek
A research lab known for strong open-weight reasoning and coding models trained at low cost. Founded 2023, based in Hangzhou. deepseek.com
Questions
Who makes DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is made by DeepSeek, founded 2023 and based in Hangzhou.
How much does DeepSeek-V3.2-Exp cost?
$0.28 per million input tokens and $0.42 per million output tokens, at its provider’s list price.
What is DeepSeek-V3.2-Exp’s context window?
160K tokens (163,840).
How does DeepSeek-V3.2-Exp rank?
#105 of 351 on LMArena’s text leaderboard, rated 1425 (95% interval 1419–1432) from 8,994 votes, as of Oct 22, 2026.
Is DeepSeek-V3.2-Exp open source?
DeepSeek-V3.2-Exp has open weights under the MIT license, downloadable from Hugging Face (deepseek-ai/DeepSeek-V3.2).