Processing or generating 1,000,000 words (approximately 1,333,000 tokens) with Llama 3.1 70B Instruct (OpenRouter) ranges from $0.1333 (cached input) to $0.5332 (full generation).
You send 1,000,000 words as prompt context, documentation, or background knowledge.
Llama 3.1 70B Instruct (OpenRouter) drafts a complete 1,000,000 words article, chapter, or code module.
800,000 words input prompt + 200,000 words output reply.
| Model | Provider | Input Cost (1,000,000 words) | Cached Input Cost | Output Cost (1,000,000 words) | Context Limit |
|---|---|---|---|---|---|
| Llama 3.1 70B Instruct (OpenRouter) (Current) | meta | $0.5332 | $0.1333 | $0.5332 | 131,072 |
| GPT-5.6 Terra | openai | $2.56 | $0.256 | $15.36 | 1,050,000 |
| GPT-5.4 Workhorse | openai | $3.20 | $0.32 | $19.20 | 256,000 |
| o3-mini | openai | $1.408 | $0.704 | $5.632 | 200,000 |
| o4-mini | openai | $1.408 | $0.352 | $5.632 | 256,000 |
| o1-mini | openai | $1.408 | $0.704 | $5.632 | 128,000 |
For Llama 3.1 70B Instruct (OpenRouter) (llama_bpe tokenizer), 1,000,000 words is approximately 1,333,000 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 1,000,000 words as input costs $0.5332 (or $0.1333 with prompt caching). Generating 1,000,000 words as output costs $0.5332. 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 $0.1333). For non-urgent asynchronous batch processing, 24-hour batch queues provide a flat 50% discount across all tokens (costing $0.2666 for input, $0.2666 for output).
At an average human reading speed of 250 words per minute, 1,000,000 words takes approximately 4000 minutes to read. In contrast, Llama 3.1 70B Instruct (OpenRouter) can process or generate this text in seconds.