How much do 1M tokens cost, and what do they buy?
Updated
1M tokens is about 750,000 English words of text, and on open models like DeepSeek V4.1 Flash it costs between $0.003 (cached input) and $0.24 (output).
The price per 1M tokens depends on the kind of token. The useful question is what a real workload costs, so this page prices three of them.
What 1M tokens costs
| Model | Input | Cached input | Output |
|---|---|---|---|
| DeepSeek V4.1 Flash | $0.06 | $0.003 | $0.24 |
| GLM 5.3 Flash | $0.12 | $0.024 | $0.40 |
| DeepSeek V4.1 Flash (fast) | $0.28 | $0.028 | $1.10 |
DeepSeek V4.1 Flash (fast) is the same model on faster serving at a higher price. See tokens per second.
What 1M tokens holds
| Content | Amount |
|---|---|
| English words | about 750,000 |
| English characters | about 3.3M |
| Chinese characters | about 1.7M |
| Novels of 90,000 words | about 8 |
The ratios come from DeepSeek's token guide (0.3 tokens per English character, 0.6 per Chinese character) and OpenAI's (100 tokens is about 75 words). Code usually packs fewer characters into each token than prose.
Three worked examples
- Coding agent session: 300 requests of about 50,000 input tokens and 800 output tokens, 90% of input cached (agents resend the same conversation every turn).
- Support assistant: 10,000 chats a month of about 3,000 input tokens and 400 output tokens, 50% of input cached (a shared system prompt and help articles).
- Summaries: 1,000 documents of about 8,000 tokens, 500-token summaries, nothing cached (each document is read once).
| Workload | Input | Output | DeepSeek V4.1 Flash | GLM 5.3 Flash |
|---|---|---|---|---|
| Coding agent session | 15M | 240K | $0.1881 | $0.60 |
| Support assistant, a month | 30M | 4M | $1.91 | $3.76 |
| Summarize 1,000 documents | 8M | 500K | $0.60 | $1.16 |
The agent session sends more tokens than the summaries but costs less, because most of its input is cached. How LLM token pricing works shows the formula behind each figure.
What changes the bill
- Cache hit rate: a stable prompt prefix moves input from the input price to the cached price.
- Reasoning: thinking tokens are billed as output. Lower the reasoning effort for simple tasks.
- Output length: set
max_tokensso a runaway answer cannot cost more than you expect. - Context size: an agent that resends 100,000 tokens of history pays for them every turn, cached or not.
Questions
How much does 1M tokens cost on DeepSeek V4.1 Flash?
$0.06 for input, $0.003 for cached input and $0.24 for output on OpenScale.
How many tokens does a coding agent use?
A long session can send tens of millions of input tokens, because each turn resends the conversation. The 300-request example on this page sends 15M input tokens and receives 240K.