veo2DeepseekVideoGenerator
🇺🇸 English

DeepSeek Suddenly Releases Its Latest Model DeepSeek-R1 and Announces Open-Sourcing Model Weights

DeepSeek-R1 has extensively utilized reinforcement learning techniques during the post-training phase, significantly enhancing the model's reasoning capabilities with minimal annotated data.

In tasks such as mathematics, coding, and natural language reasoning, its performance rivals that of OpenAI's o1 official version.

DeepSeek has open-sourced two models: DeepSeek-R1 and DeepSeek-R1-Zero, with 660B parameters.

Through model distillation, DeepSeek has also open-sourced 6 smaller models, among which the 32B and 70B models surpass OpenAI's o1-mini in multiple capabilities.

The model is licensed under the MIT License, allowing users to train other models using R1 outputs through distillation techniques.

Key Highlights

Model Performance:

  • DeepSeek-R1's performance aligns with OpenAI o1 in multiple tasks (e.g., mathematical reasoning and code generation).
  • Provides powerful reasoning capabilities and offers an API for users to call (chain-of-thought mode, set model='deepseek-reasoner').

Open-Source Content:

  • Released and open-sourced two large models (DeepSeek-R1 and DeepSeek-R1-Zero, 660B parameters).
  • Open-sourced 6 smaller models through distillation, with the 32B and 70B models outperforming OpenAI's o1-mini in multiple capabilities.

Open Licensing:

  • Standard MIT License, allowing unrestricted commercial use.
  • Explicitly supports users in training other models using DeepSeek-R1 outputs.

Products and Services:

  • DeepSeek's official website and App support inference tasks for the latest models.
  • Already available for experience on the official chat interface.

Portal: https://chat.deepseek.com/

API Pricing:

  • Cache hit: 1 RMB per million input tokens, 4 RMB if not cached.
  • Output tokens: 16 RMB per million tokens.

Detailed Price Comparison

Input API Price (Cache Hit):

  • DeepSeek-R1: 1 RMB/million tokens
  • o1-mini and o1-preview: 55 RMB/million tokens
  • o1: 11 RMB/million tokens

➡ DeepSeek-R1's price is significantly lower than other models, especially 54 times cheaper than o1-mini and o1-preview.

Input API Price (Cache Miss):

  • DeepSeek-R1: 4 RMB/million tokens
  • o1-mini and o1-preview: 110 RMB/million tokens
  • o1: 22 RMB/million tokens

➡ DeepSeek-R1 maintains a significant advantage even in cache miss scenarios, costing only 1/27 of o1-mini and o1-preview.

Output API Price:

  • DeepSeek-R1: 16 RMB/million tokens
  • o1-mini and o1-preview: 438 RMB/million tokens
  • o1: 88 RMB/million tokens

➡ In terms of output pricing, DeepSeek-R1 also leads significantly, costing only 1/27 of o1-mini and o1-preview, and is cheaper than o1.