Last updated: October 20, 2026 · Prices checked against Fish Audio's pricing docs, ElevenLabs' API pricing page (after its October 5, 2026 price cut) and Oxyy's live model pages on this date. Quality rankings from Voice Arena's public TTS leaderboard.
Disclosure: This comparison is published by the Oxyy.ai team. Oxyy is an independent gateway, not affiliated with Fish Audio or ElevenLabs.
Quick answer: Fish Audio S2 Pro is about 5× cheaper than ElevenLabs v3: $15 per 1M characters (billed as UTF-8 bytes) versus $80 per 1M characters. ElevenLabs ranks higher in blind quality tests and has more product features; Fish Audio is open-weight and had faster time-to-first-audio in Voice Arena's tests. Through Oxyy.ai you can call both on one key, at $6 (Fish S2.1-Pro) and $40 (ElevenLabs v3) per 1M characters.
| Model | Official price per 1M characters | Oxyy.ai price per 1M characters | Cost per ~1 hour of audio (Oxyy) |
|---|---|---|---|
| Fish Audio S2.1-Pro / S2-Pro | $15 (per 1M UTF-8 bytes) | $6 | ~$0.36 |
| ElevenLabs v3 | $80 | $40 | ~$2.40 |
| ElevenLabs v4 | $80 | $8.80 | ~$0.53 |
| ElevenLabs Flash v2.5 | $40 | $20 | ~$1.20 |
| Fish Audio S2.1-Pro Free | $0 | $0 (fair use) | $0 |
Official prices: Fish Audio docs and ElevenLabs API pricing page (October 2026). Oxyy prices: oxyy.ai/fish-audio and oxyy.ai/elevenlabs, October 20, 2026. "1 hour" assumes about 1,000 characters per minute of speech (60,000 characters), the rule of thumb ElevenLabs uses.
Key Takeaways
- Price: Fish Audio S2 Pro costs $15 per 1M UTF-8 bytes; ElevenLabs v3 and v4 cost $80 per 1M characters, and Flash/Turbo models cost $40.
- Quality: in Voice Arena's US-English leaderboards (October 2026), ElevenLabs v4 placed #4 in two categories; Fish S2.1 Pro placed #11 in Media & Entertainment.
- Latency: Voice Arena measured 283 ms time-to-first-audio for Fish S2.1 Pro (real-time) and 587.5 ms for ElevenLabs v3. ElevenLabs advertises ~75 ms for Flash and ~100 ms for v4 Turbo.
- Open weights: Fish S2 Pro's weights are public, so you can self-host. ElevenLabs is API-only.
- ElevenLabs cut prices 20% on October 5, 2026 and launched Eleven v4 at the same price as v3.
- One key for both: Oxyy offers 20 ElevenLabs models and 7 Fish Audio models, with Fish S2.1-Pro at $6 and ElevenLabs v3 at $40 per 1M characters.
Fish Audio vs ElevenLabs at a Glance
| Factor | Fish Audio S2 Pro / S2.1-Pro | ElevenLabs v3 |
|---|---|---|
| Official TTS price | $15 / 1M UTF-8 bytes | $80 / 1M characters |
| Billing unit | UTF-8 bytes of input text | Characters |
| Languages | 80+ (trained on 10M+ hours of audio) | 70+ (v4: 90+) |
| Max characters per request | Not published as a char limit | 5,000 (v4: 10,000) |
| Emotion / style control | Natural-language bracket tags, multi-speaker | Audio tags, multi-speaker dialogue |
| Voice cloning | Yes | Yes |
| Open weights | Yes | No |
| Time to first audio (Voice Arena) | 283 ms (S2.1 Pro real-time) | 587.5 ms |
| Speech-to-text | $0.36 / audio hour | $0.22 / hour (Scribe), $0.39 realtime |
| Rate limit model | Concurrent requests: 5 / 15 / 50 by total prepaid | Concurrency by plan |
| Extras | Voice Design ($0.01 / request) | Music, sound effects, dubbing, voice changer, agents |
Fish Audio API Pricing (Official)
| Model | Price | Notes |
|---|---|---|
s2.1-pro | $15.00 / 1M UTF-8 bytes | Recommended production model |
s2-pro | $15.00 / 1M UTF-8 bytes | Previous generation |
s2.1-pro-free | $0.00 | Same model for development and testing |
s1 | $15.00 / 1M UTF-8 bytes | Deprecated; retiring December 31, 2026 (then served as s2.1-pro) |
transcribe-1-pro / transcribe-1 | $0.36 / audio hour | Speech-to-text |
voice-design-1 | $0.01 / successful request | Generate voices from a text prompt |
Fish Audio says 1M UTF-8 bytes is roughly 180,000 English words, or about 12 hours of speech. For English, one character is one byte. For non-Latin scripts (Hindi, Bengali, Chinese, Arabic and others), one character takes 2–4 bytes, so check how your text is counted before you budget.
Fish Audio rate limits are based on concurrent requests and unlock as soon as your total prepaid amount reaches each threshold:
| Tier | Total paid | Concurrent requests |
|---|---|---|
| Starter | Under $100 | 5 |
| Elevated | $100+ | 15 |
| High Volume | $1,000+ | 50 |
| Enterprise | Custom | Custom |
ElevenLabs API Pricing (Official, After the October 2026 Cut)
On October 5, 2026, ElevenLabs cut text-to-speech API prices by 20% and launched Eleven v4. API usage is billed in US dollars at the same per-character price on every plan; the plan mainly sets your concurrency.
| Product | Price |
|---|---|
| Eleven v4 | $0.08 / 1K characters |
| Eleven v4 Turbo | $0.04 / 1K characters (~100 ms latency) |
| Eleven v3 | $0.08 / 1K characters |
| Eleven v3 Conversational | $0.04 / 1K characters |
| Multilingual v2 | $0.08 / 1K characters |
| Flash / Turbo v2.5 | $0.04 / 1K characters (~75 ms latency) |
| Scribe speech-to-text | $0.22 / hour ($0.39 realtime) |
| Eleven Music | $0.15 / minute |
Voice Quality: What Blind Tests Say
Voice Arena ranks TTS models with blind pairwise votes on sentences taken from real production calls (Bradley-Terry Elo). Here's what its US-English leaderboards showed in October 2026:
| Category | Model | Rank | Elo | Time to first audio |
|---|---|---|---|---|
| Content creation & Edu | ElevenLabs Eleven v4 | #4 | 1055 | n/a |
| Content creation & Edu | ElevenLabs Eleven v3 | #9 | 1015 | 587.5 ms |
| Media & Entertainment | ElevenLabs Eleven v4 | #4 | 1063 | n/a |
| Media & Entertainment | ElevenLabs Eleven v3 Conversational | #8 | 1030 | n/a |
| Media & Entertainment | Fish Audio S2.1 Pro (real-time) | #11 | 1008 | 283 ms |
How to read this: ElevenLabs' newest model leads Fish Audio on perceived quality in these English categories, but the confidence intervals overlap for many mid-table models, and Fish S2.1 Pro delivered its first audio about twice as fast as Eleven v3. Rankings vary by language and use case, so always test with your own scripts and voices.
Latency: Fish Audio vs ElevenLabs
- Fish Audio S2.1 Pro (real-time): 283 ms time-to-first-audio in Voice Arena's tests. Fish says S2 Pro reaches ~100 ms on a single H200 GPU with its own SGLang engine, which matters if you self-host.
- ElevenLabs v3: 587.5 ms time-to-first-audio in Voice Arena's tests. It's built for expressive, dramatic delivery rather than real-time agents.
- ElevenLabs Flash v2.5 and v4 Turbo: advertised at ~75 ms and ~100 ms. Use these, not v3, for voice agents.
Cost Comparison: Real Workloads
| Workload | Fish S2 Pro (official) | ElevenLabs v3 (official) | Fish S2.1-Pro on Oxyy | ElevenLabs v3 on Oxyy |
|---|---|---|---|---|
| 1 audiobook (~500K characters) | $7.50 | $40.00 | $3.00 | $20.00 |
| Monthly app narration (5M characters) | $75.00 | $400.00 | $30.00 | $200.00 |
| High-volume platform (50M characters) | $750.00 | $4,000.00 | $300.00 | $2,000.00 |
English text (1 character ≈ 1 byte). Oxyy prices are a current 60% "Launch Promo" on Oxyy's standard prices.
Test Both on One API Key With Oxyy
Choosing a TTS provider usually means two accounts, two keys and two SDKs. Oxyy.ai puts 20 ElevenLabs models and 7 Fish Audio models behind one OpenAI-compatible endpoint, so you can A/B test voices by changing one string.
from openai import OpenAI
import os
client = OpenAI(base_url="https://api.oxyy.ai/v1", api_key=os.environ["OXYY_API_KEY"])
for model, voice in [("s2.1-pro", "alloy"), ("eleven_v3", "Rachel")]:
with client.audio.speech.with_streaming_response.create(
model=model,
voice=voice,
input="Welcome back! Here is today's summary.",
response_format="mp3",
) as response:
response.stream_to_file(f"{model}.mp3")Oxyy TTS and Audio Prices
| Model (Oxyy ID) | Standard price | Oxyy price |
|---|---|---|
Fish Audio S2.1-Pro (s2.1-pro) | $0.015 / 1K chars | $0.006 / 1K chars |
Fish Audio S2-Pro (s2-pro) | $0.015 / 1K chars | $0.006 / 1K chars |
Fish Audio S2.1-Pro Free (s2.1-pro-free) | Free (fair use) | |
| Fish Audio Transcribe 1 / 1 Pro | $0.36 / audio hour | $0.144 / audio hour |
| Fish Voice Design 1 | $0.01 / request | $0.004 / request |
ElevenLabs Eleven v3 (eleven_v3) | $0.10 / 1K chars | $0.040 / 1K chars |
ElevenLabs Eleven v4 (eleven_v4) | $0.022 / 1K chars | $0.0088 / 1K chars |
| ElevenLabs Multilingual v2 | $0.10 / 1K chars | $0.040 / 1K chars |
| ElevenLabs Flash v2.5 | $0.05 / 1K chars | $0.020 / 1K chars |
| ElevenLabs Scribe v2 | $0.22 / hour | $0.088 / hour |
| ElevenLabs Music v2 / v2.5 | $0.15 / minute | $0.06 / minute |
"Standard price" is the list price Oxyy shows before its discount; some ElevenLabs standard prices on Oxyy predate ElevenLabs' October 5, 2026 price cut. Compared with ElevenLabs' current list prices, Oxyy's charged prices are 50% lower on v3, Multilingual v2 and Flash v2.5, 60% lower on Scribe v2 and Music, and about 89% lower on Eleven v4. Prices checked October 20, 2026.
Why use Oxyy for voice:
- One key and one balance for Fish Audio, ElevenLabs, OpenAI TTS, Gemini TTS and xAI voice models
- Free Fish S2.1-Pro tier for development
- No subscription, no minimum, credits never expire
- Free starting balance on signup, no card required
Which Should You Choose?
| If you need… | Choose |
|---|---|
| The lowest cost per hour of high-quality speech | Fish Audio S2.1-Pro (~$0.36/hour on Oxyy) |
| The top-ranked expressive English voices | ElevenLabs v4 (now the same list price as v3) |
| Real-time voice agents | ElevenLabs Flash v2.5 / v4 Turbo, or Fish S2.1 Pro real-time |
| Self-hosting or fine-tuning | Fish Audio S2 Pro (open weights) |
| Music, sound effects or dubbing in the same API | ElevenLabs |
| To test both before committing | Oxyy: one key, both providers |
Final Verdict
Fish Audio wins on price and speed. ElevenLabs wins on top-end quality and features. For audiobooks, narration and high-volume content, Fish S2.1-Pro gives you strong quality at a fraction of the cost. For premium English voices and a full audio toolkit, ElevenLabs v4 is the leader. The smart move is to test both with your own scripts, and Oxyy lets you do that on one key, for less than either provider charges directly.
Test Fish Audio and ElevenLabs on one free Oxyy key →
Related: Fish Audio API pricing · ElevenLabs API pricing · Fish Audio S2.1-Pro · Eleven v3
Frequently Asked Questions
Is Fish Audio cheaper than ElevenLabs?
Yes. Fish Audio S2 Pro costs $15 per 1M UTF-8 bytes, while ElevenLabs v3 and v4 cost $80 per 1M characters. For English text, Fish Audio is about 5 times cheaper.
How much does the ElevenLabs API cost in 2026?
After its October 5, 2026 price cut, ElevenLabs charges $0.08 per 1,000 characters for Eleven v4, v3 and Multilingual v2, and $0.04 for Flash, Turbo, v4 Turbo and v3 Conversational.
How much does the Fish Audio API cost?
Fish Audio charges $15 per 1M UTF-8 bytes for S2.1-Pro and S2-Pro, $0.36 per audio hour for transcription and $0.01 per Voice Design request. S2.1-Pro Free costs nothing.
Which has better voice quality, Fish Audio or ElevenLabs?
In Voice Arena's US-English blind tests in October 2026, ElevenLabs v4 ranked higher than Fish S2.1 Pro. Results vary by language and use case, so test with your own content.
Which is faster, Fish Audio or ElevenLabs v3?
Voice Arena measured 283 ms time-to-first-audio for Fish S2.1 Pro real-time and 587.5 ms for ElevenLabs v3. For real-time agents, ElevenLabs recommends Flash v2.5 (~75 ms) or v4 Turbo (~100 ms).
Is Fish Audio open source?
Fish Audio S2 Pro is open-weight. Its release includes model weights, fine-tuning code and an SGLang-based streaming inference engine, so you can self-host it.
Can I use Fish Audio and ElevenLabs with one API key?
Yes. Oxyy.ai offers 7 Fish Audio models and 20 ElevenLabs models through one OpenAI-compatible endpoint. Change the model name, for example s2.1-pro or eleven_v3, to switch providers.
What is the cheapest way to use ElevenLabs v3?
On Oxyy, ElevenLabs v3 currently costs $0.040 per 1,000 characters, half of ElevenLabs' $0.08 list price.
How many characters is one hour of speech?
About 60,000 characters, using ElevenLabs' rule of thumb of roughly 1,000 characters per minute of audio. Fish Audio estimates 1M bytes at about 12 hours of speech.
What are Fish Audio's API rate limits?
Fish Audio limits concurrent requests: 5 under $100 paid, 15 from $100 and 50 from $1,000, with custom enterprise limits. Tiers unlock as soon as your total prepaid amount reaches each threshold.