Cartesia vs Fish Audio
Fish Audio's numbers are fishy.
Fish Audio claims a 67% human preference rate over competitors and that they're twice as fast as Cartesia. Every independent benchmark says otherwise.
Voice quality
Independent benchmarks rate Sonic 3.6 higher
- Fish claims a 67% blind-test preference over leading competitors, and says listeners can't reliably distinguish S2.1 Pro from human across 581 published head-to-head comparisons.
- Independent benchmarks say otherwise.
Provider Voice Arena Elo
Higher is better
Latency
Cartesia is 2x faster than Fish on P90 latency
- Marketing quotes the fastest request. Users experience tails.
- Cartesia's P90 is 394ms. Fish's is 727ms. 2x faster, and not in the direction Fish claims.
Latency, P90
Lower is better
Reasons why companies choose Cartesia over Fish Audio
Human-like naturalness
Intonation, pacing, pronunciation, emotion, and audio quality that drive higher completion rates
Cartesia1282 EloFish Audio1142 EloAugust 2026
Latency (P90)
Decides whether it feels like a conversation or a phone tree.
Cartesia394msSonic 3.5Fish Audio727msJuly 2026
Emotion
Moves escalation rates and satisfaction scores.
CartesiaModel-level emotion inference from context across all languages, optional SSML tags for explicit controlFish AudioWeak model-level inference, so emotion depends on the LLM picking the right tag, SSML tags only work on 13 languagesSelf-hosting
Decides whether regulated teams can use it at all.
CartesiaVPC, on-prem, on-device, or air-gapped available all on current flagship modelsFish AudioSelf-hosting rests on open weights of older models (S2, not S2.1 Pro), with no on-device path at allEnd-to-end stack
How many vendors it takes to run one call.
CartesiaTTS, STT and Agents. Only provider ranked #1 on both speech and transcriptionFish AudioTTS-first, with an STT that doesn't compete at the frontierPVCs
How much audio you have to collect before a custom voice is usable.
CartesiaRequires just 30 minutes of audioFish AudioRequires at least an hour of audioLanguages
Comprehension and trust across a global customer base.
Cartesia40+ languages, best-in-class quality on each one, available on all tiersFish Audio83 languages claimed, but no published list and their own docs tier only 3 as top quality
Frequently asked questions
How is the quality score measured?
It comes from Artificial Analysis, an independent benchmark. Listeners hear two clips of the same line without being told which model produced either, and pick the one they prefer. The Provider Voice Arena lets every model speak in its own voices, so a voice roster counts alongside the model that speaks it. Sonic 3.6 ranks first of the 94 models rated there, and first on their Controlled Voice Arena, which gives every model the same set of cloned voices.
How consistent is Cartesia's latency under load?
Sonic 3.5 returns first audio in under 90ms, and its slowest calls sit within 43ms of its average. For a voice agent the spread matters as much as the average: a model that is usually fast but occasionally slow produces calls where the caller starts talking over it.
Can I run Cartesia on my own infrastructure?
Yes. Cartesia deploys on-prem and air-gapped, which teams in healthcare, finance, and government use to keep audio inside their own network, backed by a 99.9% uptime SLA.
How long does it take to switch?
One to two weeks for most teams. Cartesia is integrated into the major voice agent platforms and every model is on the public API, so there is no platform to move onto and nothing to unpick later.
Get started today
Talk to an expert.
Connect with a member of our team and learn how Cartesia can help you build world-class voice experiences.
Start building.
Access our models via API and bring a voice agent into production in minutes.