Audra Plus vs Essential: when 1.15× multilingual CPU is worth it — Audra

Essential is the default engine and the right one for most workloads. Plus exists for one specific gap: its 28 multilingual CPU presets and cloning-lite workflows cover voices and languages Essential does not, at a 1.15x billing multiplier (~$28.75/1M prepaid) over the same API. This post is the decision between the two.

What is actually different

Same endpoint, same request shape, same QA-gated billing — you change the model string. The differences are the voice catalog (Plus adds 28 multilingual CPU presets and light cloning workflows; voice counts and catalogs differ), the inference tier (CPU multilingual), and the price per million characters. Everything else — 50k sync limit, jobs to 500k, −16 LUFS output, X-QA-Passed — is identical.

The model swap

curl
curl https://audratalks.com/v2/speech \
  -H "Authorization: Bearer sk-audra-..." \
  -H "Content-Type: application/json" \
  -d '{"model":"audra-plus","voice":"<plus-preset-slug>","text":"Bonjour, et bienvenue...","format":"mp3"}' \
  --output plus-render.mp3

Python
import requests

resp = requests.post(
    "https://audratalks.com/v2/speech",
    headers={"Authorization": "Bearer sk-audra-..."},
    json={"model": "audra-plus", "voice": plus_slug, "text": text, "format": "mp3"},
    timeout=3600,
)
resp.raise_for_status()

JavaScript
const res = await fetch("https://audratalks.com/v2/speech", {
  method: "POST",
  headers: { Authorization: "Bearer sk-audra-...", "Content-Type": "application/json" },
  body: JSON.stringify({ model: "audra-plus", voice: plusSlug, text, format: "mp3" }),
});
if (!res.ok) throw new Error(await res.text());

A migration failure mode to avoid: swapping model without re-auditioning voices. Voice slugs do not carry between engines — your Essential pick does not exist as-is on Plus. Browse the Plus catalog (filter by engine on /compare, list with GET /v2/voices) and re-run your audition script before flipping production.

When Plus pays for itself

• You need a language/voice combination outside the Essential 72-voice catalog — the whole reason the tier exists.
• Cloning-lite workflows: approximate a reference timbre without full GPU cloning spend.
• Your volume is low enough that the 1.15x multiplier rounds to noise (a 10k-char script costs $0.25 versus $0.2875).

When it does not

If Essential has a voice you like in a language you need, Plus is 15% more money for no gain — the engines are not a quality ladder, they are different catalogs. And if you are chasing premium narration quality rather than catalog breadth, that is audra-pro (2.0x, GPU MOSS), not Plus — the Pro post covers that decision. The 15% premium only makes sense when it buys something Essential literally cannot render.

Failure mode: silent fallback

Legacy model ids (audra-core, audra-studio) still route — usually to Essential. If you think you are on Plus but forgot to change the model string, the render succeeds and bills at 1.0x with the Essential voice. Verify the model field in your logs, and compare GET /v2/usage price_per_1m_chars against the expected rate.

Full engine reference with multipliers: the four engines guide. Pricing: /pricing#engines.