DeepSeek R2
DeepSeek's second-generation reasoning model — stronger than R1 across all benchmarks at similar cost
Quick calculator
Tips to reduce cost
- →Use prompt caching to reuse repeated system prompts
- →Trim whitespace and reduce verbose instructions
- →Use a smaller model for classification or routing tasks
- →Batch async requests to get 50% discount (OpenAI/Anthropic)
- →Cache identical requests at the application layer
Similar models from DeepSeek
Compared at your current token settings
About DeepSeek R2
DeepSeek R2 is a mid-range large language model from deepseek, priced at $0.8/1M input tokens and $3.2/1M output tokens. It is 69% cheaper than the market average and best suited for advanced reasoning at low cost. The 128k context window handles long documents, extended conversations, and large code files comfortably.
As a reasoning model, DeepSeek R2 generates internal thinking tokens before responding. These are billed at the output token rate and can add 2–5x to effective output cost. For tasks requiring deep reasoning — math, complex coding, multi-step analysis — this overhead is usually justified by fewer errors and retries.
DeepSeek R2 supports prompt caching at $0.2/1M — a 75% discount on repeated input tokens. For applications with a fixed system prompt or repeated document context (RAG, chatbots, agents), enabling caching is the single highest-leverage cost optimization available.