MiniMax Speech-2.6
MiniMax's real-time TTS family: sub-250ms latency, Fluent LoRA voice cloning, and 40 languages in HD and Turbo variants.
About
MiniMax Speech-2.6 is the Chinese lab's text-to-speech family announced October 30, 2025, built for real-time voice agents rather than offline narration. It ships as two model IDs — speech-2.6-hd for fidelity and speech-2.6-turbo for low-latency conversational use — with MiniMax claiming end-to-end latency under 250 milliseconds, smarter handling of non-standard text (URLs, phone numbers, dates, amounts) without preprocessing, and a 'Fluent LoRA' pipeline that makes cloned voices sound natural across languages even from imperfect source recordings. Coverage spans 40 languages with async long-form variants for narration.
The family is widely served through third-party platforms (Novita, AI/ML API, Pipecat) as well as MiniMax's own API. One caveat for builders: MiniMax's current docs list Speech 2.6 under Legacy Audio Models, with the Speech 2.8 family now the recommended path — 2.6 still works and is widely integrated, but new projects should evaluate 2.8 first. All latency and quality figures are provider-reported; no independent TTS benchmark table exists for the family.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No published benchmarks found — latency and quality claims are provider-reported
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: MiniMax announcement Oct 30, 2025 with documented model IDs (speech-2.6-hd/turbo); third-party serving confirmed via Novita, AI/ML API, and Pipecat docs; MiniMax's own docs mark the family Legacy with Speech 2.8 current. Closed weights, API-only. Release facts corroborated across CometAPI's launch coverage, Pipecat docs, and community release notes. Strengths: real-time-first design, broad language coverage, genuine third-party adoption. Docked for: (1) latency/quality claims are provider-reported with no independent evals; (2) superseded by Speech 2.8 — new integrations should prefer the newer family.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · MiniMax announced 2025-10-30; speech-2.6-hd/turbo model IDs; third-party serving (Novita, AI/ML API, Pipecat)
- · Docked: Latency/quality claims provider-reported, no independent evals
- · Docked: Superseded by Speech 2.8 — family marked Legacy in MiniMax docs