DeepSeek V3.2 vs DeepSeek V4 Flash 0731
Compare pricing, context limits, capabilities, and benchmarks side by side.
Presets:DeepSeek V3.2 vs DeepSeek V4 FlashGPT-4o vs Claude 3.5 SonnetDeepSeek V3.2 vs GPT-4oGemini 1.5 Pro vs GPT-4oClaude 3.5 Haiku vs GPT-4o Mini
Selected Candidates:
DeepSeek V3.2
deepseek-ai • $0.25/1M
DeepSeek V4 Flash 0731
deepseek-ai • $0.14/1M
| Specification | DeepSeek V3.2 | DeepSeek V4 Flash 0731 |
|---|---|---|
Input Price Per 1M prompt tokens | $0.25 | $0.14Lowest |
Output Price Per 1M completion tokens | $0.39 | $0.28Lowest |
Context Limit Max tokens | 164k tokens | 1.0M tokensLargest |
Error Rate Hallucination % | 670.0% | 0.0%Lowest |
| Provider | deepseek-ai | deepseek-ai |
| Vision Support | No | No |
| Function Calling | Yes | Yes |
| Reasoning Mode | Standard | Yes (Reasoning Model) |
Benchmark Scores
Standardized intelligence, coding, and mathematical evaluations.
| Benchmark | DeepSeek V3.2 | DeepSeek V4 Flash 0731 |
|---|---|---|
Intelligence Index Overall AI reasoning, coding, and knowledge score. | 32.1 | N/A |
Coding Index Code generation, debugging, and software engineering. | 34.6 | N/A |
Math Index Mathematical problem solving capability. | 59 | N/A |
MMLU Pro Multi-task understanding across 14 domains. | 0.837 | N/A |
GPQA Graduate-level reasoning questions. | 0.751 | N/A |
LiveCodeBench Competitive programming problems. | 0.593 | N/A |
MATH-500 Contest and high-school math problems. | N/A | N/A |
AIME 2025 Mathematics Examination 2025. | 0.59 | N/A |
Humanity's Last Exam Frontier multi-disciplinary benchmark. | 0.105 | N/A |
DeepSeek V3.2
DeepSeek V4 Flash 0731
