AI Video & AudioStepFunStepAudio 3voice AItext to speechfree tools

StepAudio 3 Is Free Until November 8: 5 Voice and Music Models, Then $0.125 a Minute

StepFun's StepAudio 3 family, five voice and music models, is free to try until 23:59 on 8 November 2026. From 9 November, list prices apply, starting at $0.125 per session minute for realtime voice and $14 per million characters for text to speech.

Published November 6, 2026

Muhammad Usman

By Muhammad Usman · Founder & Lead Reviewer

StepAudio 3 Is Free Until November 8: 5 Voice and Music Models, Then $0.125 a Minute

Quick Answer

StepFun's StepAudio 3 family, five AI audio models covering realtime voice, transcription, text to speech, sound generation and music, is free to try until 23:59:59 on 8 November 2026, China time (UTC+8). From 00:00 on 9 November, list prices apply, with realtime voice at $0.125 per session minute and text to speech at $14 per million characters. The free window is a real deadline, not an open ended trial, so the time to test it is now.

Key takeaways

  • StepAudio 3 is a family of five models from StepFun: Realtime, ASR, TTS, Gen and Music, launched on 15 September 2026.
  • The preview models are free through 23:59:59 on 8 November 2026 (UTC+8), which is 11:59 on 8 November Eastern and 08:59 Pacific.
  • From 9 November, realtime voice costs $0.125 per session minute ($7.50 per hour) and text to speech costs $14 per million billable characters.
  • Realtime is a full duplex voice model that reasons over speech directly, with tool calling and long task execution while the conversation continues.
  • Music plans the song structure first in ABC notation, then renders stereo songs up to 5 minutes 30 seconds long.
  • There is no open weights release. StepAudio 3 is a commercial API family, not a downloadable model.
  • StepFun's own pricing page is the source for the deadline and the prices, and it is the page to check before you build anything on these rates.

What is StepAudio 3, in plain terms?

StepAudio 3 is StepFun's family of audio AI models, and it covers nearly the whole audio workflow in one place. Most AI audio tools do one job: they transcribe speech, or they read text aloud, or they make music. StepAudio 3 does all five jobs as separate models on one platform, with one API key and one billing account. If you are building a voice agent, a narration pipeline, or a content tool that needs sound, that matters, because you stop stitching together five vendors and five bills.

The five models are Realtime, for live voice conversation; ASR, for speech recognition and transcription; TTS, for text to speech; Gen, for generating voice, sound effects, ambience, background music and singing from a text description; and Music, for full songs. A Chat tier with a 256,000 token context rounds out the developer surface for text work around the audio.

The family launched on 15 September 2026, and the preview models have been free since launch. That free period now has a published end date, which is what makes this a story today rather than a quiet product page. StepFun's pricing page states the new prices take effect at 00:00 on 9 November 2026 (UTC+8), and the limited time free offer runs through 23:59:59 on 8 November 2026 (UTC+8). That is 3 days from today, 5 November.

What exactly is free, and until when?

The free offer covers the StepAudio 3 preview models, listed on StepFun's pricing page as stepaudio-3-realtime-preview, stepaudio-3-chat-preview, stepaudio-3-gen-preview and stepaudio-3-music-preview. During the free window, you can call those models without paying per minute or per character charges. StepFun also confirms a gift balance account is drawn down before a paid balance, although no numeric free grant amount is published on the official pages checked, so this article does not claim a dollar figure for starter credit.

The deadline needs care with time zones, because UTC+8 is China time, and most readers are in the US or Europe. The free window ends at 23:59:59 on Sunday 8 November 2026, UTC+8. That is 11:59 on Sunday 8 November Eastern time, and 08:59 Pacific. If you are on the US East Coast, you have until late Sunday morning. If you are on the West Coast, you have until just before 9 in the morning. Billing at list prices starts at 00:00 on Monday 9 November, UTC+8, which is noon Eastern on Sunday 8 November.

Two warnings belong here. First, StepFun states it may apply temporary throttling and adjust published limits when capacity is reached, so the free window is not a contractual guarantee of unlimited use. If many people test at once, you may be throttled. Second, the free offer is a preview, not a free tier. After 9 November, the same calls bill at list prices, and there is no published ongoing free allowance for these models in the material checked. Test now, and price your real use on the paid rates below.

What does it cost from 9 November?

The table below shows the list prices StepFun publishes for the StepAudio 3 series, effective 9 November 2026. Charges are billed in US dollars. A session minute for realtime voice counts both user speaking time and agent speaking time, with a minimum billing unit of one second, and any partial second billed as a full second.

| Model | What it does | List price from 9 November 2026 | | stepaudio-3-realtime | Live full duplex voice conversation | $0.125 per session minute ($7.50 per session hour) | | stepaudio-3-tts | Text to speech | $14.00 per 1 million billable input text characters | | Internet search (add on) | Web search inside a realtime session | $0.006 per call | | stepaudio-3-asr, gen, music | Transcription, sound generation, music | Check the live StepFun pricing page; rates for these models were not stated in the material checked |

The two prices that matter most for planning are the realtime rate and the TTS rate. At $0.125 per minute, one hour of live voice conversation costs $7.50, and a 10 minute voice agent call costs $1.25. At $14 per million characters, a 1,000 word script of about 6,000 characters costs about $0.084, or just over 8 cents. Those are list prices, and they are cheap enough that the cost of a small voice project is dominated by your time, not by the API bill. That is the honest frame: StepAudio 3 is not expensive after the free window, it is just no longer free, and your budget should use the rates above from day one.

Character counting for TTS deserves a note, because it is not one character equals one billable character for every language. One English letter counts as 1 character, one Japanese character counts as 2.5, one Chinese character counts as 3, and other characters including digits, punctuation and spaces count as 1 each. English content is the cheapest case. If you work in Chinese or Japanese, multiply accordingly before you quote a client.

What can each model actually do?

Realtime is the flagship. It is a full duplex speech model, which means you and it can talk at the same time, interrupt each other, and pause naturally, like a phone call rather than a walkie talkie exchange. It reasons directly over speech instead of transcribing first and reasoning over text, which cuts latency and keeps tone and intent in the conversation. It supports tool calling and asynchronous execution of long tasks while the conversation continues, so an agent can start a lookup, keep talking, and report back when the result arrives.

Music works differently from most AI music tools, and the difference is structural. It writes the arrangement out first, in a text shorthand called ABC notation, so harmony and song structure are decided rather than improvised, then renders stereo songs up to 5 minutes 30 seconds long. That planning step is why the output holds together over a full song instead of wandering after the first 30 seconds. If you need a backing track, an intro jingle, or a demo song with a clear verse and chorus shape, this is the model to test during the free window.

Gen is the flexible one. It generates voice, sound effects, ambience, background music and singing from a natural language description, so one prompt can produce a door creak, a rain ambience, or a sung line. ASR handles transcription, and TTS handles reading text aloud. Together, the five models cover a full audio pipeline: transcribe a meeting, summarise it, read the summary aloud, add a music bed, and ship it as a finished audio file.

How do you try it in 10 minutes?

The fastest path is the StepFun Open Platform. Create an account at platform.stepfun.ai, get an API key, and call the API with an OpenAI compatible client, using https://api.stepfun.ai/v1 as the base URL. That compatibility matters, because it means existing OpenAI client code works with a changed base URL and model name, rather than a new SDK to learn.

A sensible 10 minute test covers three calls. First, send a short paragraph to the TTS model and listen to the result, checking voice quality and pronunciation of your hardest words. Second, open a realtime session and hold a one minute conversation, interrupting once, to feel the latency and the full duplex behaviour. Third, ask the Music model for a 30 second backing track in a style you actually use, and check whether the structure holds. Save the outputs, note the settings that worked, and write down the list price of each call from the table above, so you know what the same test costs after 9 November.

If you prefer no code, StepFun also offers a Studio mode to try models straight in the browser, then wire up the API once you know what works. That is the right order: test in the browser, decide which models earn a place in your workflow, then build.

What are the catches?

Four catches deserve a plain statement. First, the free window ends on 8 November, and the paid rates start the next day. Anything you build on the assumption that StepAudio 3 is free will break on 9 November, either by billing you or by stopping. Build your pricing on the paid rates, and treat the free window as a testing budget, not a business model.

Second, there are no open weights. StepAudio 3 is a commercial API family, and no tier ships downloadable weights. If your project needs a model you can run locally or fine tune yourself, this is not it. Look at open audio models instead, and use StepAudio 3 only where an API fits.

Third, StepFun may throttle when capacity is reached, and the preview models are the ones most likely to be throttled, because they are free and popular. Do not demo a client project live on a preview model during the free window without a fallback, and do not promise a delivery date that depends on free capacity.

Fourth, the pricing page lists rates for realtime and TTS clearly, but rates for ASR, Gen and Music were not stated in the material checked for this article. Before you quote a client for a music or transcription job, open the live pricing page and confirm the rate for that exact model. Quoting a price you have not verified is how a profitable job turns into a loss.

Internal links: Category hub: https://www.topuseai.com/blog | Related live posts: https://www.topuseai.com/blog/best-free-ai-voice-tools-open-source | https://www.topuseai.com/blog/elevenlabs-starter-1-dollar-first-month-november-1-credits

Sources: StepFun Open Platform pricing page, platform.stepfun.ai/docs/en/guides/pricing/details, checked 5 November 2026; StepFun Open Platform model list, platform.stepfun.ai, checked 5 November 2026; OrcaRouter, "LFM3-Audio-2B vs StepAudio 3", 27 October 2026; ToKnow.ai, "StepAudio 3: A Voice Model That Thinks While It Speaks", 22 September 2026; Free Tier Finder promo list, checked 3 November 2026.

Frequently Asked Questions

Is StepAudio 3 really free right now?

Yes, the preview models are free through 23:59:59 on 8 November 2026 (UTC+8). From 9 November, list prices apply, starting at $0.125 per session minute for realtime voice.

When does the free StepAudio 3 offer end in US time?

At 11:59 on Sunday 8 November Eastern, and 08:59 Pacific. Billing at list prices starts at noon Eastern on Sunday 8 November.

What are the five StepAudio 3 models?

Realtime for live voice conversation, ASR for transcription, TTS for text to speech, Gen for voice, sound effects and singing, and Music for full songs.

How much does StepAudio 3 cost after the free window?

Realtime voice is $0.125 per session minute, or $7.50 per hour. Text to speech is $14 per million billable characters. Check the live pricing page for ASR, Gen and Music rates.

Can I download StepAudio 3 and run it locally?

No. StepAudio 3 is a commercial API family with no open weights release. You call it through StepFun's platform, not on your own hardware.

Does StepAudio 3 work with OpenAI client code?

Yes. It uses an OpenAI compatible API at https://api.stepfun.ai/v1, so existing OpenAI client code works with a changed base URL and model name.