Speech 2.8 key stats
- Latency (measured)
- 325 ms1
- Vapi streaming benchmark (50 trials per model) (checked 2026-06-10) Measured on the MiniMax realtime turbo tier (June 2026 run, configured with the speech-2.5-turbo-preview request id); median of 50 sequential live streaming trials including network RTT.
- platform.minimax.io/docs/api-reference/speech-t2a-http (checked 2026-06-10) Vendor lists 40+ languages for the current speech generation.
- platform.minimax.io/docs/guides/pricing-paygo (checked 2026-07-23) Turbo tier $60 per 1M characters pay-as-you-go (speech-2.8-turbo; HD bills $100); the arena clips are the 2.8 generation, turbo tier (matches the measured realtime latency).
- platform.minimax.io/docs/guides/models-intro (checked 2026-07-23, confidence: medium) Speech 2.8 family (speech-2.8-hd / speech-2.8-turbo). The arena clips are this generation, turbo tier, per MiniMax; the Speech-02 series preceded it in 2025-04 and Speech 2.5 in 2025-08.
Background
MiniMax's speech models moved fast through 2025 and into 2026, with the Speech-02 series arriving in April 2025, Speech 2.5 following in August, and the Speech 2.8 generation current since early 2026. The current generation supports more than 40 languages and clones a voice from roughly six to ten seconds of reference audio, using a learnable speaker encoder that needs no transcript. MiniMax is widely regarded as the strongest text to speech provider for Chinese, and its recent generations brought English accuracy and rhythm up alongside it.
Sources: platform.minimax.io
At a glance
The arena clips on this Index were generated with the Speech 2.8 generation, turbo tier, the realtime tier we also measured for latency. In our 50 trial streaming benchmark it returned first audio in a median of 325 ms.
Sources: platform.minimax.io
Position in the rankings
Standings as of Jul 28, 2026, 01:54 UTC
Frequently asked questions
- How is Speech 2.8 tested on the Humanness Index™?
- Listeners hear Speech 2.8 against another model in a blind head to head round, both voices reading the same customer support prompt from the same cloned source voice, and they pick whichever sounds more human. Its Humanness score derives purely from those votes.
- Which MiniMax generation do the arena clips use?
- The clips were generated with the Speech 2.8 generation, turbo tier, per MiniMax, the realtime tier we also measured for latency (325 ms median TTFB). This entry was labeled Speech 2.5 until July 2026, when MiniMax confirmed the benchmarked generation was 2.8.
- Was the HD or Turbo variant benchmarked?
- Turbo. Both the clips and the 325 ms latency figure come from the realtime turbo tier. The blind evaluation itself runs on 20 English customer support prompts, while MiniMax advertises more than 40 languages of capability for the generation.
Find the most human-sounding voice for your agent.
Compare the models in blind tests, read the methodology, or get in touch.
Build a TTS model? Add yours to the Index.
