Inference that costs 90% less.

Text to speech, chat, and batch inference on the best open models. Works with the OpenAI and ElevenLabs SDKs you already use. Change one line and keep everything else.

Join the waitlist View docs

pip install openai

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["MSERVE_API_KEY"],

− ~~# base_url defaults to api.openai.com~~+

)

audio = client.audio.speech.create(
    model="kokoro-82m", voice="aria",
    input="Your order is on its way.",
)

From $0.06 / 1M charsText to speech

From $0.15 / 1M tokensBatch inference

From $0.30 / 1M tokensChat, input

OpenAI SDKChange base_url, nothing else

ElevenLabs SDKChange base_url, nothing else

Apache 2.0 · MITCommercial use included

Products

Speech, chat, and batch inference on open models.

Voice

Natural voices from the best open models. Clone a voice from a 10-second clip. Stream as you generate.

from $0.06 / 1M charsExplore Voice

Real-time

Chat and completions on open models, through the OpenAI API you already use.

from $0.30 / 1M tokensExplore Real-time

Batch

Run millions of requests overnight. Drop-in for the OpenAI Batch API.

from $0.15 / 1M tokensExplore Batch

Speech that streams.

Audio starts before the sentence ends.

Explore Voice

Chat behind the OpenAI API.

Streaming replies from open models, same SDK.

userSummarize this call in one line.

assistantCustomer asked to move delivery to Thursday; confirmed and sent the updated receipt.

Explore Real-time

Batch at our lowest price.

One file in, results within your window.

input.jsonl0 / 120 done

Explore Batch

Your bill, after.

Voice cloning, per million characters. ElevenLabs list price against Chatterbox.

Per 1M charactersElevenLabsmserve

Why teams switch.

Nothing to rewrite.

Keep your SDK, your code, and your prompts. Swap the base URL.

Prices you can see.

One public price per model. No tiers, no minimums, no quotes.

Credits that never expire.

Prepaid, billed per character or per token. No monthly plan.

Models you can build a business on.

Apache 2.0 and MIT licensed. Commercial use included.

Models and pricing

Model	Product	Best for	License	Price
`Kokoro 82M`	Voice	Preset voices	Apache-2.0	$3.00/ 1M chars	/ 1M chars
`Chatterbox`	Voice	Voice cloning	MIT	$6.00/ 1M chars	/ 1M chars
`Qwen3-TTS`	Voice	10 languages	Apache-2.0	$8.00/ 1M chars	/ 1M chars
`Chatterbox Voice Conversion`	Voice	Voice conversion	MIT	$0.06/ minute	/ minute
`Qwen3.8 27B`	Real-time	Chat	Apache-2.0	in$0.30out$2.50/ 1M tokens	/ 1M tokens
`Qwen3.8 27B (Batch)`	Batch	Batch jobs	Apache-2.0	24 h$0.22flex$0.15/ 1M tokens	/ 1M tokens

FAQ

Does it work with the OpenAI SDK?

Yes. Chat, completions, and batch work with the official openai libraries. Set base_url and your key.

Does it work with the ElevenLabs SDK?

Yes. Text to speech, streaming, and voice cloning work with the official elevenlabs library.

How is it this cheap?

We run open models on efficient hardware and pass the savings on. Every price is public.

Is there a free tier?

Every new account starts with free credits. Your invite states the amount.

When can I get access?

Voice opens first. Join the waitlist and we email you the day your product is ready.

Can I use the voices commercially?

Yes. Every model we serve is Apache 2.0 or MIT licensed.

Start building for less.

One email when your product opens. Nothing else.