Update DeepSeek · Released
DeepSeek V3.2
What is DeepSeek V3.2?
DeepSeek V3.2 is a Update AI model from DeepSeek, released on February 12, 2026. It has a 1M tokens context window, costs $0.27 per million input tokens, and scores 1424 on Chatbot Arena Elo. It is available as open source.
Open-weight MoE with a 1M+ token context window and strong coding.
Major update with 10x context window expansion to over 1 million tokens
Key features
- 1M+ token context window (10x expansion)
- Improved reasoning capabilities
- Open source release
- Cost-effective inference
How much does DeepSeek V3.2 cost?
| Metric | Price |
|---|---|
| input per mtok | $0.27 |
| output per mtok | $1.10 |
| cached input per mtok | $0.07 |
| cache read per mtok | $0.50 |
Pricing data verified .
What context window does DeepSeek V3.2 have?
- Context window
- 1M tokens
- Max output
- 8K
- Input modalities
- text
- Output modalities
- text
How does DeepSeek V3.2 score on benchmarks?
| Benchmark | Score |
|---|---|
| MMLU | 90.1% |
| HumanEval | 92.5% |
| MATH | 85.6% |
| GPQA | 68.4% |
| LiveCodeBench | 86.2% |
| MMLU-Pro | 86.2% |
| GPQA Diamond | 84.0% |
| HLE | 22.2% |
| SciCode | 38.9% |
| TAU2-bench | 90.6% |
| TerminalBench-Hard | 35.6% |
| IF-Bench | 60.7% |
| LiveCodeBench Reasoning | 65.0% |
| AA Intelligence Index | 41.7 |
| Chatbot Arena Elo | 1424 |
Benchmark scores verified .
What's new
- 10x context window expansion (128K → 1M+ tokens)
- Sliding-window attention for long-context throughput
- Improved chain-of-thought reasoning
- Native FP8 inference support
How does DeepSeek V3.2 compare to DeepSeek-V3?
- TAU2-bench: 22.8% → 90.6% (+67.8pt)
- Input pricing: $0.27 → $0.27
- Context window: 128K tokens → 1M tokens