Murf API at $0.01 a minute vs Sume TTS per character

On paper Murf is cheaper: its API page lists $0.01 per minute, while Sume's text-to-speech works out to roughly 4 to 5 cents per minute of finished speech. The two are billed in different units (minutes versus characters), so the comparison only holds once you pick a speaking rate, and this post shows the arithmetic.

The Murf figures below come from its page as read on 2026-10-02. The Sume figures come from the Sume TTS Router catalog fields and the API reference. Treat the Sume number as arithmetic, not a quote: the live price for your request is what the job reserves.

What does Murf list for its API?

Murf's page names two speech models, Falcon (a fast model, 55 ms model inference and sub-130 ms end-to-end latency) and Gen 2 (precise voice controls, 99.38% pronunciation accuracy). It lists 150+ voices across 35 languages, and a flat price of $0.01 per minute with no tiered complexity pricing.

The same page says the API handles up to 10,000 concurrent calls at the same latency, and that Murf is SOC 2 Type II, ISO 27001, GDPR and HIPAA compliant. It also lists Dubbing, Voice Changer and Translation APIs. It does not say what a minute means for a request (generated audio or request time), so confirm that with Murf before budgeting.

How does Sume bill text to speech?

Sume bills per transcript character, spaces and punctuation included, up to 20,000 characters per request. The TTS Router catalog (GET /v1/tts-router/models) carries the numbers: every Sonic row has list_usd_micros_per_character: 38 and billable_margin: 1.25, with the formula list x 1.25 -> ceil usd cents.

That is $0.038 per 1,000 characters at list and about $0.0475 per 1,000 after margin, before the cent rounding the formula applies. Confirm the live number in GET /v1/catalog.

What does a minute of speech cost on Sume?

Pick a rate. This post assumes 150 words per minute and 6 characters per word including the space, so 900 characters per minute. That is an assumption for illustration; your script's characters per minute will differ with language and pacing.

At 900 characters per minute, one minute of speech is about $0.043 and one hour is about $2.57. Against Murf's listed $0.01, Sume is roughly four times as expensive per minute of audio on this basis.

What do you get for the difference?

A per-minute rate is simplest when you only need audio. Sume's value is in the pieces around it. TTS jobs can return word timings (timestamps: { words: true }) and sentence segments that drive captions and timeline slides. Voices can be chosen by avatar, so a spoken line matches the avatar it will appear on. Output defaults to MP3 at 44.1 kHz and 128 kbps, with WAV and raw containers and sample rates from 8 kHz to 48 kHz.

Sume also bills all of this on one USD wallet next to music, transcription, audio detach and timeline audio jobs, so a voiceover, a music bed and a render land on one invoice.

What Sume does not do: it does not match a flat $0.01 per minute, and its TTS endpoint prices by character, not by minute.

How do I call it?

Submit with an Idempotency-Key, wait up to 30 seconds with mode: sync, and fall back to polling the job if it has not finished. The model below is a real catalog id; the voice comes from an avatar in your workspace.

curl -X POST https://api.sume.com/v1/tts-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: murf-compare-001" \
  -d '{
    "model": "sonic-latest",
    "transcript": "Welcome back. Today we are comparing two ways to pay for narration.",
    "avatar_handle": "@my.avatar",
    "mode": "sync",
    "wait_timeout_seconds": 30
  }'

Which should I pick?

If your only requirement is cheapest minutes of generic narration, Murf's listed rate wins and you should verify the voice quality yourself. If the narration feeds a Sume timeline, an avatar render or caption alignment, the per-character price buys timing data and one billing surface. Run your real script through both and compare cost per finished minute, not per page.