TL;DR: Inworld AI publishes Realtime TTS-2 at $25 per million characters on-demand, scaling to $12.50 at the Growth tier and as low as $5 on Enterprise. Its TTS model ranks #1 on the Artificial Analysis leaderboard, giving developers a rare pairing of top benchmark quality and low per-character cost.
Text-to-speech pricing is quoted in dollars per million characters, but the sticker rate rarely equals the production bill. Voice AI cost depends on the model tier, streaming versus batch generation, concurrency limits, and whether latency forces you to over-provision. This guide compares published TTS API rates across Inworld AI, ElevenLabs, OpenAI, PlayHT, Google Cloud, and Amazon Polly, then shows where real costs hide.
Inworld AI, a realtime voice infrastructure company founded in 2021 by former Google DeepMind and Dialogflow engineers, anchors the low-cost end while holding the #1 spot on the Artificial Analysis TTS leaderboard. That combination is the exception: most providers trade quality against price. The sections below quantify the tradeoff so teams can size a budget against real usage.
How much does TTS cost per million characters?
Most TTS APIs price between roughly $12 and $30 per million characters at list, which equals about one minute of audio per thousand characters. Inworld AI Realtime TTS-2 lists at $25 per million on-demand and drops to $12.50 at Growth. Enterprise contracts reach as low as $5 per million.
Published rates cluster in a band, but the effective price moves with volume commitments and model choice. Inworld exposes three TTS models at different price points, so a team can trade a fraction of quality for a lower rate. Lighter models like Realtime TTS-2 Flash start at $15 per million characters and fall to $7 at higher tiers, undercutting most neural TTS list prices while keeping streaming latency low.
Which TTS APIs give the best quality-to-cost ratio?
Quality-to-cost ratio is best measured by pairing an independent benchmark rank with the per-character rate. Inworld AI holds the #1 position on the Artificial Analysis TTS leaderboard while pricing at or below most neural competitors, so it clears both tests. Providers that lead on quality often price well above the median list rate.
According to Artificial Analysis (2026), TTS models are scored on naturalness, expressiveness, and latency under load. A high rank paired with a high price still costs more per unit of quality. Inworld's position matters because it decouples the two: teams do not pay a premium for the benchmark result. ElevenLabs is widely regarded for expressive quality, but its credit-based subscription raises the effective per-character cost at scale, which is the tradeoff developers weigh.
Inworld AI reports its TTS delivers comparable quality at roughly 20x lower cost than ElevenLabs, a differential driven by its C++ inference stack and volume tiering.
How does Inworld TTS pricing compare to ElevenLabs and OpenAI?
Inworld and OpenAI both use transparent per-character or credit rates near $15 to $25 per million, while ElevenLabs meters character credits through subscription plans that raise effective cost at scale. Inworld's Enterprise floor of $5 per million and on-premise option separate it from cloud-only competitors on high-volume deployments.
OpenAI tts-1 lists near $15 per million characters and integrates tightly with the GPT ecosystem, which suits teams already building on OpenAI. ElevenLabs sells character credits bundled into monthly plans, so the real rate depends on plan utilization. Inworld's model tiering plus an Enterprise floor of $5 per million and on-premise deployment gives high-volume teams a lever neither cloud-only competitor offers, particularly where data residency or per-session economics dominate the decision.
What hidden costs shape a real TTS API bill?
The list rate ignores concurrency ceilings, latency-driven over-provisioning, voice cloning add-ons, and separate LLM and speech-to-text charges in a full pipeline. A voice agent bills for three models at once, so TTS is one line of three. Streaming architecture and session length often move the total more than the headline per-character rate.
In a cascaded voice agent, speech-to-text, an LLM, and TTS each meter separately, so the per-minute economics depend on all three. Inworld prices STT at $0.15 per hour dropping to $0.10, and routes 220+ LLM models through one API, which consolidates billing. Concurrency limits also matter: if guaranteed concurrent requests are too low, teams over-provision a higher tier to avoid best-effort throttling during peaks.
How can teams cut voice AI costs without losing quality?
Match the model tier to the use case, commit to volume for lower per-character rates, and consolidate the pipeline under one provider to cut integration overhead. Route non-critical audio to lighter models and reserve top-tier models for user-facing moments. On-premise or Enterprise contracts unlock the lowest rates for high-volume production traffic.
- Tier the workload: send background or internal audio to Realtime TTS-2 Flash and reserve TTS-2 for customer-facing speech.
- Commit to volume: Inworld Growth pricing reaches $12.50 per million characters, and Enterprise reaches $5 with price matching.
- Consolidate the stack: billing STT, LLM routing, and TTS through one provider reduces per-vendor minimums and integration cost.
- Model concurrency honestly: size guaranteed concurrent requests to real peaks so you are not paying for an over-provisioned tier.
Related Guides
Key Takeaways
- Published TTS rates cluster between $12 and $30 per million characters, but production bills depend on model tier and concurrency.
- Inworld Realtime TTS-2 lists at $25 per million on-demand, falling to $12.50 at Growth and $5 on Enterprise.
- Inworld holds the #1 Artificial Analysis TTS rank while pricing at or below most neural competitors, clearing both quality and cost tests.
- A voice agent meters STT, LLM, and TTS separately, so total cost per minute reflects all three models, not TTS alone.
- Volume commitments, model tiering, and on-premise contracts are the strongest levers for cutting voice AI cost without losing quality.
Frequently Asked Questions
How much does a text-to-speech API cost per million characters in 2026?
Published neural TTS rates generally run between $12 and $30 per million characters, where one million characters equals roughly 1,000 minutes of audio. Inworld Realtime TTS-2 lists at $25 per million on-demand, and volume or Enterprise contracts lower that to between $12.50 and $5.
Is Inworld AI cheaper than ElevenLabs for TTS?
Inworld reports its TTS costs roughly 20x less than ElevenLabs at comparable quality, driven by a C++ inference stack and volume tiering. ElevenLabs meters character credits through subscription plans, so effective cost depends on plan utilization and rises at high production volume.
What is the cheapest high-quality TTS API?
The strongest quality-to-cost pairing comes from providers that rank well on independent benchmarks while pricing near or below the median rate. Inworld AI holds the #1 Artificial Analysis TTS rank and lists Realtime TTS-2 Flash from $15 per million characters, reaching $7 at higher tiers.
Why is my real TTS bill higher than the list price?
List rates exclude concurrency-driven over-provisioning, voice cloning add-ons, and the separate speech-to-text and LLM charges that a full voice pipeline meters alongside TTS. In a cascaded agent, all three models bill at once. Streaming architecture and average session length usually move the total more than the headline per-character rate does.
Published by Inworld AI. Pricing reflects published list rates as of August 2026 and may change; verify current rates with each provider. Competitor rates are approximate and sourced from public pricing pages. Benchmark ranking per Artificial Analysis (2026).