Models

Pick a model with the model field of each request. Leave it out and vakyn-max answers. You pay for input tokens only, from your prepaid balance; output is free on every model.

Current models

Prices are in US dollars. Input tokens are the state and the questions you send, as counted in usage.input_tokens. Calls that end in an error cost nothing.

Choosing a model

Set model in the body of POST /v1/systemone. Every question in the request goes to that model. The model field of the response names the model that answered.

curl -s "https://api.vakyn.com/v1/systemone" \  -H "Authorization: Bearer $TYPESAFE_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "model": "vakyn-fast",    "state": "My order #4471 arrived with a cracked screen. I would like a replacement, not a refund.",    "questions": {      "damaged": {        "type": "noul",        "instructions": "Does the customer report a damaged item?"      }    }  }'

With the SDKs you can also set TYPESAFE_DEFAULT_MODEL in the environment, so code that never names a model picks yours. A name the API does not know gets a 400 with model_not_found:

{  "detail": {    "error_type": "model_not_found",    "message": "Unknown model \"vakyn-mini\". Use one of: vakyn-max, vakyn-fast. See https://vakyn.com/docs/models."  }}

When to pick which

Not sure? Start with vakyn-max. Then run a sample of your own cases on vakyn-fast and compare the answers and how sure they are. If they are good enough for your threshold, switch and pay less. The Confidence page shows how to set that threshold.

Aliases

Some names point at a model. Use them when you want code to follow the default.

jev-latest is the name code written for TypeSafe's Jev API sends by default, so that code works unchanged. The response names the model the alias pointed at, not the alias.

Limits

Both models have the same limits.

A request over a token limit gets a 400 with max_tokens_exceeded. It is never cut short: every answer is computed on all of what you sent. To work with longer documents, see Long documents. The full list of errors is in the API reference.

Rate limits

Each API key can send 60 requests per second. Short bursts up to that number pass. Above it, the API answers 429 with a Retry-After header: wait that many seconds and send again. Errors like this cost nothing.

Need more? Tell us what you send and how often, and we can raise the limit for your organization.

Languages

The models are tested on English. Other languages are untested: they may work, but measure them on your own cases before you rely on them.

Listing models

GET /v1/models lists every model with its price, its limits and whether it can be called right now (available), plus the default and the aliases. The SDKs call it with models.list().

curl -s "https://api.vakyn.com/v1/models" \  -H "Authorization: Bearer $TYPESAFE_API_KEY"

{  "default": "vakyn-max",  "aliases": {    "vakyn-latest": "vakyn-max",    "jev-latest": "vakyn-max"  },  "models": [    {      "id": "vakyn-max",      "name": "VAKYN MAX",      "available": true,      "price_per_billion_input_tokens": 30,      "price_per_billion_output_tokens": 0,      "max_tokens_state_and_longest_question": 32768,      "max_tokens_request": 65536,      "input": "text"    },    {      "id": "vakyn-fast",      "name": "VAKYN Fast",      "available": true,      "price_per_billion_input_tokens": 23.31,      "price_per_billion_output_tokens": 0,      "max_tokens_state_and_longest_question": 32768,      "max_tokens_request": 65536,      "input": "text"    }  ]}