Sesame CSM-1B
Sesame's open conversational speech model: context-aware dialogue TTS under Apache 2.0, the base behind the viral Maya voice demo.
About
CSM-1B is the open release of Sesame's Conversational Speech Model line, published March 13, 2025 by the lab behind the viral Maya voice demo. Rather than plain text-to-speech, CSM generates Mimi RVQ audio codes from combined text and audio inputs: a Llama-style transformer backbone plus a smaller audio decoder lets it produce dialogue speech that responds naturally to the preceding conversation. The weights sit on Hugging Face as sesame/csm-1b under Apache 2.0, natively supported in Hugging Face Transformers from version 4.52.1, with ~14.7k GitHub stars making it one of the most-watched open speech releases.
Practical notes: the base checkpoint isn't tuned to any specific voice and has limited non-English ability; it needs a CUDA GPU, Python 3.10, and gated access to both the CSM-1B and Llama-3.2-1B downloads. Sesame's terms prohibit impersonation and deception, and the larger CSM-3B/8B models that power its commercial product remain closed. Sesame has since raised a $250M Series B (October 2025) and launched Maya and Miles as full agents — the open 1B remains the community's self-hosted conversational-voice base.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
No published benchmarks found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (metrics tier, metrics-v1). Evidence: Sesame release Mar 13, 2025 with Apache-2.0 weights on HF (sesame/csm-1b); native Transformers support from 4.52.1; ~14.7k GitHub stars; lab funded ($250M Series B Oct 2025, Sequoia/Spark). Open weights. Release facts corroborated across the official repo, HF model page, and BuilderAI's model profile. Strengths: genuinely open conversational-speech architecture, strong community traction, permissive license. Docked for: (1) no published benchmarks — quality claims rest on the viral demo and community reports; (2) gated HF access and the Llama-3.2-1B dependency add friction versus fully open drops.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Sesame release 2025-03-13; Apache-2.0 weights on HF; native Transformers 4.52.1 support; ~14.7k GitHub stars; $250M Series B Oct 2025
- · Docked: No published benchmarks — quality rests on demo and community reports
- · Docked: Gated HF access + Llama-3.2-1B dependency add friction