Processing or generating 50,000 words (approximately 66,650 tokens) with DeepSeek R1 (Reasoner) ranges from $0.009331 (cached input) to $0.146 (full generation).
You send 50,000 words as prompt context, documentation, or background knowledge.
DeepSeek R1 (Reasoner) drafts a complete 50,000 words article, chapter, or code module.
40,000 words input prompt + 10,000 words output reply.
| Model | Provider | Input Cost (50,000 words) | Cached Input Cost | Output Cost (50,000 words) | Context Limit |
|---|---|---|---|---|---|
| DeepSeek R1 (Reasoner) (Current) | deepseek | $0.0367 | $0.009331 | $0.146 | 64,000 |
| GPT-5.6 Terra | openai | $0.128 | $0.0128 | $0.768 | 1,050,000 |
| GPT-5.4 Workhorse | openai | $0.16 | $0.016 | $0.96 | 256,000 |
| o3-mini | openai | $0.0704 | $0.0352 | $0.2816 | 200,000 |
| o4-mini | openai | $0.0704 | $0.0176 | $0.2816 | 256,000 |
| o1-mini | openai | $0.0704 | $0.0352 | $0.2816 | 128,000 |
For DeepSeek R1 (Reasoner) (deepseek_bpe tokenizer), 50,000 words is approximately 66,650 tokens (an average ratio of 1.33 tokens per word in English). Code, technical vocabulary, and non-English scripts will have higher token densities.
Sending 50,000 words as input costs $0.0367 (or $0.009331 with prompt caching). Generating 50,000 words as output costs $0.146. Output tokens are more expensive because autoregressive token generation requires significantly more computation per token.
Yes! If your input text is part of a repeated context, prompt caching saves 50% to 90% (costing $0.009331). For non-urgent asynchronous batch processing, 24-hour batch queues provide a flat 50% discount across all tokens (costing $0.0183 for input, $0.073 for output).
At an average human reading speed of 250 words per minute, 50,000 words takes approximately 200 minutes to read. In contrast, DeepSeek R1 (Reasoner) can process or generate this text in seconds.