Back to Models
DeepSeek: DeepSeek V4 Flash 0731
Deepseektext8.9 / 10 Overall Rating
DeepSeek V4 Flash 0731 is a 284B total (13B active) sparse Mixture-of-Experts model designed for high-throughput coding, reasoning, and agentic workflows. It offers an expansive 1.3M-token context window and output generation up to 943K tokens at roughly 1/50th the input cost of GPT-4o and Claude 3.5 Sonnet. While ideal for massive document processing and rapid agent execution, its lightweight active parameter count limits high-level reasoning depth on complex edge cases.
Context Window
1311K
Knowledge Cutoff
2025
Max Output
16K
License
Proprietary
Where it Excels (Pros)
- Massive 1.3M token context window
- Extremely low API pricing
- Fast inference via sparse MoE
- Optimized for coding and agents
Limitations (Cons)
- Text-only with no multimodal support
- Lower reasoning quality than GPT-4o
- Lacks Claude's deep writing nuance
Benchmark Breakdown
Reasoning (MMLU)82%
Coding (HumanEval)78%
Mathematics (MATH)76%
Long-Context Retrieval95%
Pricing Matrix
Input cost / 1M tokens$0.06
Output cost / 1M tokens$0.12
Compare DeepSeek: DeepSeek V4 Flash 0731 with other frontier models: