OpenRouter pricing: DeepSeek V4.1 Flash at $0.14/$0.42 per 1M tokens
The author explains why output tokens cost 3–5x input using OpenRouter prices, listing DeepSeek V4.1 Flash at $0.14/$0.42 versus Opus 5.5 at $4/$20.
Ever wondered why output tokens cost 3 to 5x more than input tokens? Some current @OpenRouter prices per 1M tokens (input / output): Claude Sonnet 5.5 → $2 / $10 GPT-6.1 Sol → $2 / $10 Gemini 4 Argon → $2 / $10 Claude Opus 5.5 → $4 / $20 DeepSeek V4.1 Flash → $0.14 / $0.42 When a model reads your prompt, it processes the whole thing in one pass, all at once. It writes the answer step by step, a token or a few at a time, and each step has to wait for the one before it. So each output token keeps the GPU busy much longer than an input token. A long context also makes each of those passes slower, because the model has more past text to look back at on every step. What this means when you build: - Reasoning tokens bill as output. If a model thinks for 5,000 tokens and writes a 200-token answer, you pay for 5,200 output tokens. - Token for token, a shorter answer saves 3 to 5x more than a shorter prompt. But agents often send 10 to 100x more input than output, so looks at your traces to see what your workloads are actually doing. - Cached input is the cheapest option (on Sonnet 5.5, $0.10 per 1M vs $2 for fresh input) when the start of your prompt stays byte-for-byte the same. Put your system prompt and tools first, and put anything that changes at the end. - Batch APIs cut both input and output prices in half if you can wait for the answer.