ElevenLabs vs Cartesia vs Fish Audio: voice APIs for phone agents and apps

The short answer before the detail

If you had to pick in 30 seconds: Cartesia for a genuinely latency-sensitive live voice agent, ElevenLabs for the broadest toolkit if you're willing to pay more, Fish Audio if cost per credit and a usable free tier matter most and your use case is asynchronous rather than live. The detail below covers real current pricing and where each one is actually built to win.

What are ElevenLabs, Cartesia and Fish Audio?

ElevenLabs is the broadest of the three: text to speech, voice cloning, conversational agents, dubbing and speech to text through one credit-based account. Cartesia is a voice AI company built specifically around low-latency, real-time generation, the kind of near-instant response a live phone call or voice agent needs to feel natural rather than laggy. Fish Audio is a text-to-speech and voice-cloning platform positioned as a lower-cost, generous-free-tier alternative, with pay-as-you-go API pricing separate from its subscription plans.

What makes these three genuinely different companies, not just different prices

It's easy to treat these three as interchangeable voice APIs that just happen to charge different amounts, but each was actually built around a different core problem. ElevenLabs was built around voice quality and breadth, as many voice-related capabilities as possible under one roof. Cartesia was built around speed, the engineering problem of generating speech with minimal delay for real-time systems. Fish Audio was built around accessibility and cost, making voice generation and cloning available to more people at a lower price point, including a free tier generous enough to actually use commercially. Those different starting points explain the pricing and feature differences better than comparing line items alone.

Latency, price and voice quality

Latency is Cartesia's whole reason for existing, its product and pricing pages are built around real-time agent use specifically, including a dedicated per-minute voice-agent rate (roughly 3 cents a minute at its Scale tier) separate from its credit-based text to speech pricing. ElevenLabs also supports low-latency generation, with its Business plan specifically discounting that use case to roughly 5 cents a minute, but latency isn't the sole focus of its product the way it is for Cartesia. Fish Audio doesn't market itself primarily around latency, its positioning is more about affordable, high-volume generation with a large free allowance. Why an AI phone agent replies slowly explains what else adds delay.

On price, Fish Audio's free tier is the most generous of the three, 8,000 credits a month (roughly 7 minutes of generation) with commercial use allowed even on the free plan, a meaningful difference from ElevenLabs' free tier, which has no commercial license at all. Cartesia's free tier gives 20,000 credits a month. ElevenLabs' free tier gives 10,000 credits a month but reserves commercial rights for paid plans. Open source ElevenLabs alternatives covers the run-it-yourself route.

Provider	Free tier	Entry paid plan	Real-time/agent focus
ElevenLabs	10,000 credits/month, no commercial license	Starter: $5/month annual, 30,000 credits	Supported, discounted on Business plan
Cartesia	20,000 credits/month	Pro: $5/month, 100,000 credits	Core focus, dedicated per-minute agent pricing
Fish Audio	8,000 credits/month, commercial use allowed	Plus: $11/month annual ($15 monthly), 250,000 credits	Not the primary focus

Best for phone agents

If you're building a live phone or voice agent where every fraction of a second of delay is noticeable to the caller, Cartesia is purpose-built for exactly that problem, and its pricing structure reflects it with a dedicated per-minute agent rate. ElevenLabs is a solid second choice if you also want its broader agent-building tools and don't want a second vendor for cloning or dubbing. Fish Audio is the weakest fit of the three specifically for real-time agents, it's a stronger choice for batch or asynchronous generation (videos, audiobooks, pre-recorded content) where latency doesn't matter and cost per credit does.

A worked example: a customer support voice agent handling 500 calls a month

Say your business runs a voice agent handling 500 support calls a month, averaging 4 minutes each, 2,000 minutes of live call audio total. On Cartesia's Scale tier agent rate (roughly 3 cents a minute), that's about $60 a month in call-time cost, on top of the $299-a-month flat Scale subscription. On ElevenLabs' Business plan discounted agent rate (roughly 5 cents a minute), the same 2,000 minutes costs roughly $100, on top of the $825-a-month Business subscription (though Business includes considerably more general-purpose credits than Cartesia's equivalent tier, so the comparison isn't purely about the per-minute rate). Fish Audio doesn't publish a comparable dedicated agent per-minute rate, since real-time agent work isn't its primary focus, so it's a weaker fit for this specific scenario regardless of its lower general credit pricing.

Setup and integration effort

All three offer developer APIs, but the amount of surrounding work differs. Cartesia's product is built end to end around low-latency streaming, so integrating it into a live call system tends to need less custom latency-optimization work on your end, since the platform itself is engineered for that specific constraint. ElevenLabs' broader toolkit means more available features (cloning, dubbing, a full agent framework) but potentially more decisions to make while integrating. Fish Audio's simpler, lower-cost API suits a more straightforward batch-generation integration, generate audio, store it, serve it, rather than a live, latency-critical call flow. ElevenLabs versus OpenAI text to speech compares a fourth API.

The verdict

If you're unsure which of the three fits, start by asking whether your use case is live or asynchronous. A live phone call or real-time voice interaction narrows the field to Cartesia or ElevenLabs' own agent product immediately, since Fish Audio isn't built around that constraint. An asynchronous use case, pre-recorded video, audiobooks, batch content, opens the field to all three, and cost per credit becomes the more relevant tiebreaker once latency is off the table. Choose based on the job, not the name recognition. Cartesia for a genuinely latency-sensitive live agent. ElevenLabs if you want the broadest toolkit and are willing to pay more for it. Fish Audio if your use case is asynchronous and cost per credit is the deciding factor, its free tier alone (8,000 credits with commercial rights) is worth testing before committing to a paid plan on any of the three.

Need something completely custom?

Everything above is a path to get it running quick and cheap. But sometimes the honest answer is that none of those quite fit, and what you actually need is something custom that you own, built uniquely for your business from the ground up, with AI thoughtfully designed into the parts of it that your processes actually benefit from. An asset that grows in value, not a generic, one-size-fits-all boilerplate that's an expense with a lock-in. This is a complete bespoke build, and it's something I do very well. Get in touch to know more.

How does Blur Studio approach voice agent builds?

Picking the right voice API is only one decision in a real voice agent build, the call handling, the logic behind what the agent says and does, and what happens when it hands off to a human all matter just as much. Tell me what you're building and I'll help you pick the right engine and build the rest around it properly.