MiniMax Speech ✓

Positioned as the top tier for tonal nuance and timbre similarity ("Perfecting Tonal Nuances. Maximizing Timbre Similarity.").

Available as speech-2.8-hd and speech-2.8-turbo in the official T2A model list.

Fits best for premium voice likeness where subtle delivery details matter most.

Same T2A interface as the rest, so swapping is mainly a quality-latency choice, not an integration rewrite.

Designed to work in real-time flows, including WebSocket streaming TTS usage.

Optimized for ultra-low latency plus stronger naturalness for voice-agent style interactions.

Available as speech-2.6-hd and speech-2.6-turbo in the official T2A model list.

Framed as a comprehensive upgrade for agent scenarios: faster response, better handling, and more human-like speech.

Great "interactive default" when you want responsiveness without sacrificing expressiveness.

Works with the same T2A constraints and tooling (for example up to 10,000 chars per request).

The baseline generation in the family for high-quality, expressive TTS, and still a strong reference point.

Available as speech-02-hd and speech-02-turbo in the official T2A model list.

Highlighted by MiniMax as earning top placements on external TTS leaderboards (listed as Speech-02-HD).

Good fit for voiceovers (HD) vs real-time (Turbo) style selection within the same core model line.

Uses the same T2A API surface (stateless calls, standard voice controls like pitch and speed).

The foundational tier that establishes the "generative speech model" approach for the lineup.

Available as speech-01-hd and speech-01-turbo in the official T2A model list.

Positioned by MiniMax around high fidelity plus controllability for expressive speech synthesis.

Supports the same T2A request shape and practical constraints (for example 10,000 chars per request).

Best used as the dependable baseline tier when you want consistent output behavior across many voice styles.