Gpt-5-nano seems slow, is there a quicker model that accepts parameters?

API
1

I am currently using gpt-5-nano to built an agentic flow but it is very slow - as in 22 seconds slow. Is there another model I can use that still uses parameters?

One that would still work with the structure of my calls
e.g.

const RATING_SCHEMA = {

type: ‘object’,

additionalProperties: false,

properties: {

scored_options: {

  type: 'array',

  items: {

    type: 'object',

    additionalProperties: false,

    properties: {

      mode: { type: 'string' },

      label: { type: 'string' },

      score: {

        type: 'number',

        description: 'Relevance percentage from 0-100 (integer).'

      },

      rationale: {

        type: 'string',

        description: 'Short explanation of the score.'

      }

    },

    required: \['mode', 'label', 'score'\]

  }

}

},

required: [‘scored_options’]

};

const { parsed } = await callOpenAIFunction({

  openai,

  name: 'rate_router_options',

  description: 'Assign relevance scores to the router options.',

  parameters: RATING_SCHEMA,

  model: 'gpt-5-nano',

  temperature: 0,

  messages: \[

    { role: 'system', content: instructions },

    { role: 'user', content: JSON.stringify(payload) }

  \]

});

Thanks,

2

Despite the name, gpt-5-nano generates an extremely large quantity of “thinking” you pay for. Yet benchmarks lower.

The model cannot accept a temperature other than a default of “1”, meaning “no effect”. Inspect your function and what is actually ran, as its odd that your function could even work if it is not actually stripping this parameter out on unsupported models. Additionally, “parameters” is a poor function signature name for passing an enforced response format into a “call AI” function.

Try gpt-5-mini. Try passing the API parameter reasoning.effort = "low" to transition to seen output faster.

If you like even better quality, have the AI reason about “rationale” BEFORE it makes its score. Otherwise the AI is merely trying to justify what it saw it wrote earlier.

The model switch alone might get you speed, quality, without much more cost.