Sesame Ends Its Free Preview: $20 a Month for Two Hours of Talk Time a Day, $100 for Ten

Sesame has taken its voice assistant out of preview and put a meter on it. The company now sells three tiers built around talk time rather than messages or tokens: Free keeps 30 minutes of conversation a day, Plus costs $20 a month for two hours a day, and Pro costs $100 a month for ten hours a day. The tiers also scale a weekly credit allowance, at four times the free allowance on Plus and twenty times on Pro.

The overage rules are where the pricing gets interesting. Extra talk time costs $5 for one hour or $20 for five, and credit packs sell at $5 per thousand and $20 per five thousand, with the packs never expiring. An extra hour at $5 works out to roughly 8.3 cents a minute, which is expensive for a chat and cheap for a product a business puts in front of its own customers. Because add-on packs are spent after the plan allowance, the marginal hour is the number to model, not the headline price.

What is being metered is a voice-first agent app rather than a voice model. Sesame's assistant runs named personas with memory that survives between sessions, and it can search, take notes and set reminders during a call, then hand back a summary when the call ends. It has been in public preview since May 2026, which is why the paid tiers read as the end of a trial rather than the launch of a new product.

There is a supply-side gap worth knowing about before anyone benchmarks the thing. Sesame publishes a model, CSM-1B, with open weights on Hugging Face under an Apache 2.0 license, but that is not the voice in the app. The production system exposes no API, no model identifier and no independent arena score, and the research release has not moved since 2025. Buyers comparing Sesame against a vendor with published model cards are therefore comparing a product against components, which is a different decision.

The move follows a broader repricing of consumer voice. Meta began selling outbound calling on its Muse assistant on November 12 to US subscribers, at $20 a month for its Power plan and $100 a month for its Maximum plan, metered in weekly tokens rather than talk time. On the cost side, hosted voice inference has been getting cheaper through the year, with per-minute agent pricing at the low end of the market now measured in cents rather than tens of cents. Sesame's decision to charge for time, not tokens, puts the meter on the input that actually scales with how much value a customer gets.

That is unusual for consumer AI, which has settled into flat monthly fees almost everywhere. A talk-time allowance prices the one resource that grows with both usage and compute at the same time, and it gives the company a lever if heavy users turn out to be unprofitable. It also sets an implicit ceiling on how the product gets used: at ten hours a day on Pro, the plan is priced for someone treating the assistant as a working tool, not a novelty.

The open questions are whether the allowance model survives contact with competitors who will likely offer flat pricing, whether Sesame ever publishes an API and model card that let enterprises evaluate the voice on its merits, and whether talk time is the right unit at all once agents start doing work in the background rather than holding a conversation. For now the company has made a clear bet: the value of a voice assistant is the time it spends talking, and that time now has a price.