DeepSeek V3.2 reasoner is the thinking-mode variant of DeepSeek V3.2, delivering reasoning performance competitive with o3 and Claude Opus at roughly 1/10th the price. It supports up to 64K output tokens and is the strongest cheap option for math and coding reasoning.
Pricing Breakdown
| Volume | Input Cost | Output Cost | Combined (50/50) |
|---|---|---|---|
| 1,000 tokens | $0.0003 | $0.0004 | $0.0003 |
| 10,000 tokens | $0.0028 | $0.0042 | $0.0035 |
| 100,000 tokens | $0.0280 | $0.0420 | $0.0350 |
| 1,000,000 tokens | $0.2800 | $0.4200 | $0.3500 |
Strengths
- ✓o-series-class reasoning at a fraction of the price
- ✓Up to 64K output tokens for long chain-of-thought
- ✓Open weights
- ✓Cache-hit pricing available
Weaknesses
- ✗Verbose reasoning increases effective output cost
- ✗128K context vs. 200K on US competitors
Compare DeepSeek V3.2 (Reasoner) With
Frequently Asked Questions
How much does DeepSeek V3.2 (Reasoner) cost?
DeepSeek V3.2 (Reasoner) costs $0.28/1M per 1M input tokens and $0.42/1M per 1M output tokens.
What is DeepSeek V3.2 (Reasoner)'s context window?
DeepSeek V3.2 (Reasoner) has a context window of 128K tokens, which means it can process up to 128,000 tokens in a single request.
Is DeepSeek V3.2 (Reasoner) open source?
Yes, DeepSeek V3.2 (Reasoner) is open source. The model weights are publicly available and can be self-hosted.
What is DeepSeek V3.2 (Reasoner) best used for?
DeepSeek V3.2 (Reasoner) is best suited for: reasoning, math, coding, research. o-series-class reasoning at a fraction of the price.
What is DeepSeek V3.2 (Reasoner)'s ELO score?
DeepSeek V3.2 (Reasoner) has an ELO score of 1385 on the LMSYS Chatbot Arena leaderboard, placing it among top-tier models.