DeepSeek V3.2
Chatdeepseek/deepseek-v3.2DeepSeek V3.2 is DeepSeek's latest general-purpose chat model, released on 2025-12-01, building on the series' instruction-following and coding abilities. Pre-trained on 15 trillion tokens, it delivers an excellent cost-performance ratio for everyday development and production workloads. The model supports tool use (function calling) and prompt caching, which cuts the cost of repeated context. Context window: 128K tokens, output: 32K. Pricing is $0.29/M input tokens and $0.43/M output tokens, with cached input reads at $0.06/M. Available via OpenAI and Anthropic protocols through Ofox.
Providers
/v1/chat/completions/v1/responsesCode Examples
from openai import OpenAIclient = OpenAI(base_url="https://api.ofox.run/v1",api_key="YOUR_OFOX_API_KEY",)response = client.chat.completions.create(model="deepseek/deepseek-v3.2",messages=[{"role": "user", "content": "Hello!"}],)print(response.choices[0].message.content)
Benchmarks
DeepSeek V3.2 scores 1425 in the Overall category of the LMArena text leaderboard (style control), ranking #87 of 374 models based on 47,248 human preference votes (updated 2026-07-12).
| Category | Arena Score | 95% CI | Rank | Votes |
|---|---|---|---|---|
| Overall | 1422–1429 | #87 of 374 | 47,248 | |
| Hard Prompts | 1443–1452 | #85 of 374 | 26,180 | |
| Coding | 1463–1476 | #89 of 369 | 10,613 | |
| Math | 1418–1440 | #74 of 362 | 3,004 | |
| Creative Writing | 1393–1409 | #80 of 372 | 6,678 | |
| Instruction Following | 1415–1427 | #77 of 374 | 12,919 | |
| Chinese | 1450–1476 | #90 of 344 | 2,441 |
Source: LMArena · CC BY 4.0 · Updated 2026-07-12 · Methodology ↗ · Ranks compare models within each category of the LMArena text leaderboard (style control). Scores come from third-party human preference evaluations, not from OFOX.