OpenAI TTS
OpenAI · tts · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
tts-1 verifies at ≈$0.011/min — but gpt-4o-mini-tts bills $12/1M audio output tokens with no published token-to-audio-second rate: the per-minute cost is unverifiable.
#11 of 16 in Tts
ToolScore breakdown — why 58/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸tts-1 verifies at ≈$0.011/min — the same value band as Fish Audio's API
- ▸API access with no separate license tier — commercial terms are the API terms
- ▸The plain API integration story: one key, standard SDKs
Where it costs you
- ▸gpt-4o-mini-tts bills $12/1M audio output tokens with no published token-to-audio-second rate — per-minute cost unverifiable
- ▸No AA Speech Arena row in our snapshot — no independent quality reference
- ▸No latency figure, no self-host path
Hidden traps — the stuff the pricing page doesn't say
gpt-4o-mini-tts bills $12/1M audio output tokens, and no token-to-audio-second conversion is published. Unlike Google (25 tokens/sec), there is nothing to multiply: the per-minute cost is a guess, and our hub doesn't publish guesses — the cell stays "—" until a rate or a measurement appears.
Know the limits before you pay: read the traps above, then decide with eyes open.
Visit official siteAffiliate link where available — never affects the score.
- official OpenAI API pricing — tts-1 and gpt-4o-mini-tts audio output token billing (official, checked 2026-11-12) verified
- internal ToolScore TTS hub — paid table: tts-1 ≈$0.011/min, gpt-4o-mini-tts "—" (verify pending) verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
The odd one out: edits existing audio by text instruction — fix a word, change the emotion — plus ze
Apache 2.0 two-speaker dialogue in a single pass — [S1]/[S2] interruptions and reactions without sti
≈$0.011/min on the API with Elo 1274 quality — and the local weights are CC-BY-NC, so commercial use
OpenAI TTS review
In-depth analysis · last updated 2026-11-12
The unverifiable minute
OpenAI is the row our TTS hub built its token-billing rule around. tts-1 — the older model — verifies at ≈$0.011/min, fine. gpt-4o-mini-tts, the current one, bills $12/1M audio output tokens, and publishes no token-to-audio-second rate. Nobody outside OpenAI can compute what a finished minute costs. Our hub's position is principle, not oversight: token billing can hide the minute, and until the rate appears, the cell stays "—" — verify pending.
The economics, half-verified
What's checkable: $12/1M audio output tokens. What's missing: the conversion — Google publishes 25 tokens/sec and becomes arithmetic; OpenAI publishes nothing and stays a riddle. If gpt-4o-mini-tts ran at Google's rate it would cost ≈$0.018/min; if it ran at half that rate, ≈$0.036/min. The spread between those guesses is the budget risk, and we won't pick a number and call it a price.
Real limits
PAYG — no tiers or published ceilings in our snapshot, no separate commercial license tier (the API terms are the terms). No latency figure, no self-host path, and — notably — no row in the AA Speech Arena top-8 we use as quality reference: the blind-vote board our whole hub leans on simply doesn't rank it.
How the pillars land
Price/Value 6.5: tts-1's verified ≈$0.011/min carries the pillar; the flagship model's unverifiable price caps it. Quotas 6: clean PAYG, nothing published to game. Latency 5.5 — no latency fact exists in our snapshot; this is an editorial estimate and marked as one. Training use & voice cloning 5.5: no policy facts — near-neutral pending audit. API & self-host 6.5: API only, standard integration. Feature set 6: no AA arena row in our snapshot — the quality reference is absent, and we score the absence. Track record 7 — editorial; a live, billing, documented API. Ease 7: one key, standard SDKs, one unverifiable number. Pricing transparency 4: the entire verdict — audio-token billing without a published rate in a category where the rate is the price. Support & docs the neutral 5 — no support facts in our snapshot.
Verdict — PRICING UNVERIFIABLE, 58/100
Use tts-1 at its verified ≈$0.011/min if the OpenAI integration is already in your stack and the older voice is good enough. For gpt-4o-mini-tts, either wait for a published rate or measure your own tokens-per-minute and publish it — the number that belongs in our table is one only OpenAI or your own meter can provide. Until then, Google publishes its rate, Simba publishes its chars, Cartesia publishes its credits: three vendors sell minutes you can actually budget.