DeepSeek V3.2
Current · Released: December 1, 2025
- Context window
- 128.000 tokens
- Max output
- 8.192 tokens
- Price input / 1M tokens
- $0.27
- Price output / 1M tokens
- $0.42
Capabilities
- Tool use
- Thinking
- Open source
Benchmark scores
- MMLU-Pro
- 85.0
- AIME 2025
- 93.1
- HMMT 2025
- 83.6
- Codeforces
- 2,121.0
- GPQA Diamond
- 82.4
- LiveCodeBench
- 83.3
- SWE-bench Verified
- 73.1
- Humanity's Last Exam
- 19.8
Notes
Open source Chinese model. Excellent price-to-performance ratio, up to 90% cheaper than Western competitors. Mixture-of-experts architecture with 671B parameters, 37B active per token.