Rime

Our full take

Rime is a voice engine, not a full agent platform. You do not log in and build a phone bot. You take Rime’s text-to-speech and plug it into something else (Vapi, LiveKit, Pipecat) that handles the call. What Rime sells is the voice itself, and specifically a voice that gets the hard words right. Its own line is “built for calls you can’t afford to get wrong,” and that is the honest centre of the pitch: pronunciation accuracy on names, medications, account numbers and addresses, the kind of detail that quietly loses a customer when an agent fumbles it.

The pricing is usage-based and billed per character, so there is no single all-in number to quote, and it got simpler (and dearer at the entry point) in mid-2026. The redesigned pricing page now shows one figure: starting at $0.05 per 1,000 characters, with volume discounts above it. The old per-model rates (Mist at $0.03, Arcana at $0.04) no longer appear anywhere on the page, so the cheap Mist rate is gone as a published price. The models themselves are all still current in the docs: Mist v3 for low latency, Arcana v3 for expressive multilingual work, and Coda, the flagship, which the docs now recommend migrating Arcana traffic to. New accounts start on Starter with 3,000 minutes free and 20 concurrent generations, then pay as they go. Enterprise is a custom per-character rate.

Because every comparison on this site works in per-minute terms, here is the workings. A minute of spoken agent audio is roughly 900 to 1,000 characters. In a real phone call the agent only talks for about half the time, the caller has the other half, so a conversation-minute generates closer to 450 to 500 characters of speech. At $0.05 per 1,000 that is about $0.023 to 0.025 of voice per minute. Add your own language model (call it $0.01 a minute) and a phone line such as Twilio (about $0.014 a minute) and a realistic all-in sits around $0.05 to 0.10 a minute, a touch above where it sat when the $0.03 Mist rate was published. Two things to be clear about: that all-in is a model, not a rate Rime quotes, and the language-model and telephony parts are not Rime charges, they are what you pay other suppliers in the stack Rime sits inside.

One wrinkle worth flagging. “Starting at” is doing real work in that headline: it implies volume discounts the page does not spell out. The page itself has settled since we caught it mid-redesign with placeholder FAQ text on 11 July 2026 (the real FAQ copy was in place at our 23 July re-check), but the advice stands: if you are negotiating, get the current rate card in writing rather than trusting the single published figure, and ask directly whether a cheaper Mist-class rate still exists at volume.

On voices and reach, Rime advertises a library of 300-plus voices, with Arcana v3 carrying 94 flagship voices on its own. The language list is short though: seven languages (English, Spanish, French, Portuguese, German, Japanese and Hebrew, per the docs as of 19 May 2026). A caveat on that: Rime’s pricing FAQ and its July 2026 funding announcement both now claim 600-plus voices across 50-plus languages, figures its own docs do not yet reflect. We keep the docs numbers until the two agree, and if language breadth is your deciding factor, ask Rime for the current list in writing. On the docs’ account, if you need to sound native across twenty markets this is not the engine for that yet. Where it concentrates its effort is English-language realism and getting domain-specific words right, which is a deliberate trade rather than a gap.

Compliance is where Rime is interesting for the regulated buyer, with a caveat. The pricing page states that the Enterprise tier includes a BAA (HIPAA) and SOC 2 reports, alongside on-prem and VPC deployment options. Rime also leans hard into healthcare in its marketing and runs on Oracle Cloud Infrastructure for that market. So the HIPAA claim is sourced and ticked here. SOC 2 is mentioned as “reports” on the Enterprise tier, but we have not yet linked a primary certification letter or stated a Type 1 versus Type 2 level, so both SOC 2 boxes stay unticked until we see the paperwork. GDPR is not mentioned on the pricing page at all, so it is unticked too. If you are in healthcare or finance, treat the HIPAA line as the starting point of a conversation with their sales team, not the finish line, and get the BAA and the SOC 2 report scope in writing before you build.

On the company itself, Rime announced a $24 million Series A on 15 July 2026, led by M13 with Twilio Ventures, Corazon Capital, Unusual Ventures and Cadenza participating, and named Rafael Valle as chief scientist. The release pitches Rime toward an “enterprise-ready speech-to-speech model” and claims a scale of roughly 100 million calls a month, both the vendor’s own framing. Funding is not a product feature, but for a voice engine you wire deep into a call stack, a freshly funded supplier is worth knowing about.

A few practical limits. Rime brings no speech-to-text, no language model and no telephony of its own, so the headline per-character price genuinely is just the voice. That is fine if you are a developer wiring up a stack, and real setup work if you are a non-technical buyer expecting to plug in and go. There is no documented self-serve affiliate or partner programme either, so the link from this page is the plain website, not a tracked one. And the ease-of-use score reflects that Rime is a building block, not a finished product: you need a framework around it.

My read: Rime earns a place on the shortlist when the deciding factor is pronunciation accuracy on high-value, often regulated calls, and you are already building your own agent rather than buying an all-in platform. The per-character pricing is still reasonable at $0.05 per 1,000, though the published $0.03 budget rate is gone and Cartesia now undercuts it on narration cost. The HIPAA path is real, and Mist is built for the low-latency phone use case. Look elsewhere if you need a large multilingual voice library, voice work in languages outside the seven, or a platform that hands you the phone line and the language model in one bill.

The 1 to 10 scores and the latency figure on this page are an editorial preview, our provisional read to get the framework in place, not a measured result. We have not run Rime through our own listening or call tests yet, so there is no Voxrater latency figure here. The pricing, voice and compliance detail is sourced from Rime’s pricing, docs and homepage plus the LiveKit, Pipecat and Vapi integration pages, first captured 2026-05-31 and most recently re-captured 2026-07-23.