DeepSeek
DeepSeek V4 Pro API model, version DeepSeek-V4-Pro-0813, with thinking and non-thinking modes.
1M ctxPeak: USD 1.32 / 1M input tok; USD 3.96 / 1M output tok
Introducing DeepSeek V4.1 FlashDeepSeekReleased September 10, 2026
DeepSeek V4.1 Flash
DeepSeek current native-multimodal Flash model with an asymmetric 552B-parameter MoE architecture.
multimodal-llmOpen WeightsPeak: USD 0.30 / 1M input tok; USD 1.20 / 1M output tokContext: 1M (1,000,000 tokens)
Technical Specifications
Architecture Type
Causal Encoder-Decoder MoE
Total Parameters
552B
Active Parameters (MoE)
8B input / 16B output
Context Window
1M (1,000,000 tokens)
Max Output Tokens
384K (384,000 tokens)
Knowledge Cutoff
Frontier (Current)
Supported Modalities
text, vision, reasoning
License & Access
Open Weights
Deep Architectural Overview
DeepSeek V4.1 Flash is called with deepseek-flash. It supports thinking and non-thinking modes, native visual understanding, a 1M-token context window, and a 384K maximum output.
Strengths & Considerations
Core Strengths
- Native visual understanding
- 1M-token context
- 8B active input and 16B active output parameters
Token & API Pricing
Input Tokens (1M)$0.30
Output Tokens (1M)$1.20
Cached Input (1M)$0.0060
Pricing is verified directly against DeepSeek's developer documentation and API rate sheets.
Similar & Alternative Models
Explore other frontier models from DeepSeek and comparable reasoning engines.