Pricing
What each model costs, per million tokens, and how to estimate the cost of a real workload.
Updated on Aug 10, 2026
Text models are billed by the number of tokens processed. Input tokens include the instructions, messages and other content you send; output tokens correspond to what the model generates. The values below are in dollars per 1 million tokens.
Not everything is billed this way: images are charged per generated unit, audio per minute, and embeddings on input only. Each case has its own section on this page.
| Model | Input | Output |
|---|---|---|
hinow/himax | US$ 2.26 | US$ 9.00 |
hinow/hinova | US$ 0.69 | US$ 7.49 |
hinow/higenesis | US$ 0.22 | US$ 0.98 |
Check the catalog in production
Prices and availability may change. Use GET https://api.hinow.ai/v1/models to read the current value and the modalities enabled for your account.
| Model | Input | Output | Specialty |
|---|---|---|---|
hinow/hivision | US$ 1.00 | US$ 3.00 | Dedicated image input |
hinow/hicode | US$ 0.49 | US$ 5.48 | Code generation and transformation |
hinow/hivision is the right choice when the input is predominantly visual. hinow/hicode is a specialized option for code automation. Compare both with the general-purpose HINOW models when the task combines text, business rules and code.
| Model | Price | Billing |
|---|---|---|
hinow/himegia | US$ 0.082 per image | Per image generated or edited |
hinow/himegia bills per image, not per token: US$ 0.082 — the same amount to create and to edit, with a short or long prompt, in any aspect ratio and format. Every POST https://api.hinow.ai/v1/images response carries the cost field with the exact amount for the call. The guided walkthrough is in Image generation.
| Model | Price | Billing |
|---|---|---|
hinow/hiembed | US$ 0.05 per 1M tokens | Input only |
hinow/hiembed bills input only: US$ 0.05 per million tokens. There is no output price because what comes back is a vector, not generated text — in the response, usage carries prompt_tokens and no completion_tokens.
The cost leans on indexing, not on querying: embedding 2 million tokens costs US$ 0.10, once; after that each query costs only the embedding of the question — a 10-token sentence comes to US$ 0.0000005. The guided walkthrough is in Embeddings.
| Model | Price | Billing |
|---|---|---|
hinow/hivox | US$ 0.05 per minute | Proportional to audio length |
hinow/hivox bills by duration, not by token: US$ 0.05 per minute of audio, the same amount for any format and any spoken language. The response carries duration in seconds, and that is what the bill comes from, proportionally — 8.54 seconds cost US$ 0.0071.
In the order of magnitude that matters when planning: a one-hour meeting costs US$ 3.00; a hundred five-minute support calls, US$ 25.00. The guided walkthrough is in Audio transcription.
Availability may vary by account. Query the catalog at runtime to obtain the identifiers, modalities, endpoints and prices currently available:
curl https://api.hinow.ai/v1/models \
-H "Authorization: Bearer $HINOW_API_KEY"{
"data": [
{
"id": "hinow/himax",
"category": ["text_to_text"],
"cost": { "type": "mtoken", "input": 2.26, "output": 9 },
"endpoint": "/v1/chat/completions"
}
]
}The current price comes from the API
GET https://api.hinow.ai/v1/models returns cost.input and cost.output per million tokens for each model. Use these fields in calculators, usage dashboards and model selectors to avoid an outdated copy of the table.
The cost of a call is input tokens × input price + output tokens × output price. Chat Completions responses include usage with both counts. Log this field to measure the real cost of each workflow:
"usage": { "prompt_tokens": 132, "completion_tokens": 60, "total_tokens": 192 }In this example, the call to hinow/hinova costs 132 × 0.69/1M + 60 × 7.49/1M, approximately US$ 0.00054. With streaming, consume the stream up to the final event to record the usage returned by the API.
- Use the model that fits the task. Start with HiNova, measure quality and move to HiMax or HiGenesis when the result justifies the change.
- Define the expected length. An instruction such as "answer in up to 60 words" guides the style and reduces unnecessarily long outputs. Validate the limit in the application when it is mandatory.
- Trim the history. In long conversations, every call resends previous messages. Summarize or remove turns that no longer change the answer.
max_tokensas a safety net. It cuts mid-sentence (finish_reason: "length"), so use it as an operational ceiling, not as the only way to control length.
Which model for which task
The strengths of each model and where each one performs best.