gpt-4o-mini-tts at $12 per 1M characters vs a Sume TTS job

OpenAI's pricing page lists gpt-4o-mini-tts at $12 per 1M characters of audio output, plus $0.60 per 1M text input tokens. That is about $0.012 per 1,000 characters before the token line. Sume's TTS router bills about $0.0475 per 1,000 characters, so on raw price OpenAI is roughly four times cheaper (read 2026-10-03).

The per-character rates side by side

All OpenAI figures come from the OpenAI pricing page. The Sume figure is the catalog formula for the Sonic models: the upstream list price of $38 per 1M credits at one credit per character, multiplied by 1.25, then rounded up to whole cents per job.

A 5,000-character article

At 5,000 characters, gpt-4o-mini-tts costs about $0.06 in audio output, plus a small token charge for the input text. The same text through Sume costs 5 x $0.0475 = $0.2375, rounded up to $0.24 for the job.

• OpenAI: about $0.06 plus input tokens.
• Sume: $0.24 for the job after rounding up to the cent.

What the extra money buys

Price per character is not the only line item. The OpenAI guide describes 13 built-in voices, six output formats and an instructions parameter for tone, and it requires you to tell end users the voice is AI-generated.

Sume's TTS router takes a transcript of 1 to 20,000 characters per request, selects a voice by avatar id, avatar handle or voice id, and can return word timestamps with sentence segmentation. It fails a request that would run past 1,200 seconds of audio with tts_duration_exceeded. If you already keep Sume voice clones, the same ids work here, which saves a second voice vendor.

When to pick which

If you only need generic narration at the lowest cost, the OpenAI rate wins. If the audio feeds a Sume video, caption or timeline job, the shared job envelope and hosted media URLs matter more than a few cents. The 1,200-second limit post covers how to split long chapters.