Checked against the OpenRouter public models feed. Prices are copied, not estimated.

The Best Model ForWhich AI model to use, from ten labs.

The best AI model for voice

Speak with a model, or ask questions about a recording.

Short answer

Start with GPT Audio ($2.50 in · $10.00 out per million tokens). GPT Audio is the reviewed model when the product must talk back. The feed lists audio in and audio out. Listen to it on your script before you trust the token rate. A cheaper stop is GPT Audio Mini ($0.60 in · $2.40 out).

Quality

GPT Audio

OpenAI · $2.50 in · $10.00 out

GPT Audio is the reviewed model when the product must talk back. The feed lists audio in and audio out. Listen to it on your script before you trust the token rate.

GPT Audio is listed at $2.50 input and $10.00 output per million tokens, with a 128,000-token context window.

Balanced

GPT Audio Mini

OpenAI · $0.60 in · $2.40 out

Same listed shape as GPT Audio, lower published token rate. This is the budget model that still speaks.

GPT Audio Mini is listed at $0.60 input and $2.40 output per million tokens, with a 128,000-token context window.

Cheapest sensible

Voxtral Small 24B 2507

Mistral · $0.10 in · $0.30 out

Voxtral Small is the cheap open-weight reader for a recording. You lose spoken output. If all you need is a transcript, a dedicated transcription model may be cheaper still, and this guide does not rank transcription accuracy.

Voxtral Small 24B 2507 is listed at $0.10 input and $0.30 output per million tokens, with a 32,768-token context window.

Also on the feed

These match the factual limits of the task (modality, tools, context) and are not the reviewed picks. A row marked new has no guide note.

Transcription only — accuracy not reviewed

These rows list transcription as the output. They are cheaper when you do not need a conversation. The Best Model For has not ranked their word-error rates.

Questions

Why might the bill not match the token price?

Some audio fields on the feed are priced per token and some look like a different unit. The model page shows those raw fields. A zero token price is not a promise that the audio feature is free.

Which model only transcribes?

Rows whose output modality is transcription. They are listed on this page as a factual list. The Best Model For has not reviewed their accuracy.