MiniMax Audio is a text-to-speech and voice-cloning platform from MiniMax, the Shanghai-based AI company also known for its Hailuo video models and M-series LLMs. It targets creators, marketers and developers who need multilingual voiceovers or a cloned narrator voice, and it ships two ways: a consumer web app at minimax.io/audio with monthly subscriptions from $5, and a pay-per-character developer API on MiniMax's Open Platform. The current flagship model, Speech 2.8, focuses on natural, less robotic-sounding delivery.
Who builds it
MiniMax Group Inc. was founded in Shanghai in December 2021 by Yan Junjie, a former SenseTime computer-vision researcher, and listed on the Hong Kong Stock Exchange in January 2026. The workflow is the same on web and API: enter a script, pick one of 300+ pre-built voices or clone your own from a short sample, tune emotion, pitch and speed, then export. Developers reach the same engine via MiniMax's Open Platform using HTTP/WebSocket endpoints, an async endpoint for long text, and dedicated Voice Clone/Voice Design endpoints.
Core features
- Speech 2.8 (January 23, 2026), the flagship model, adds "Native Sound Tags" that model natural fillers ("um," "uh") and breathing, plus improved cross-lingual handling that reduces accent bleed, starting with Mandarin-Japanese pairs.
- Instant voice cloning from as little as 10 seconds of source audio (the API accepts 10 seconds to 5 minutes, plus an optional under-8-second sample for extra fidelity).
- 300+ studio-quality pre-built voices spanning 50+ languages and accents, organized by gender, age, accent and style, with adjustable pitch, speed and emotion.
- Speech 2.6 (October 2025), the prior flagship, targets voice agents with sub-250ms latency and automatic reading of URLs, emails, phone numbers, dates and money without manual preprocessing.
- A developer API exposing synchronous HTTP/WebSocket TTS, an async endpoint for long documents, and separate Voice Clone/Voice Design endpoints billed independently from generation.
- Tiered voice-library capacity: subscription plans include rising "voice slots" for storing cloned voices, from 10 on the entry plan up to 800 on the top published plan.
Pricing
- Five Audio subscription tiers plus a custom option: Starter $5/month (100,000 audio points), Standard $30/month (300,000), Pro $99/month (1,100,000), Scale $249/month (3,300,000), and Business $999/month (20,000,000 points, 800 voice slots); Custom via sales.
- Quarterly billing saves 10%, annual saves 20%, on every tier.
- No permanent free plan is published; new visitors get a small number of free trial generations (MiniMax's voice generator page showed "5 free trials remaining" during our check) before a paid tier is required.
- Separate pay-as-you-go API pricing: speech-2.8-turbo costs $60 per million characters, speech-2.8-hd costs $100 per million characters; cloning is billed per voice ($1.50 Rapid Voice Cloning, $3.00 Voice Design).
- Always check MiniMax's current pricing page, since these tiers and rates have changed more than once in the past year.
Who it's for
Creators, marketers and indie developers needing occasional multilingual voiceovers or a one-off cloned voice will likely find the $5-30/month tiers enough once trial generations run out. Teams building voice agents, embedding TTS into a product, or needing large voice libraries and higher rate limits should look at Pro/Scale/Business or the pay-as-you-go API, which unlock more voice slots, higher rate limits and (on Business) priority access to new models.