Cartesia's Startup plan costs $49 a month for 1.25 million credits, and it costs $49 whether a month of calls uses all of them or a tenth. A credit is one text to speech character, so the plan is an allowance, and a voice agent's real spend depends on how much of it gets used.
Bland Speech bills characters as they're spoken, from a wallet balance, at one standard rate. That changes how a migration reads: the streaming code maps across with a few edits, and the larger question is whether a per-character meter or a monthly allowance fits a team's call volume better.
Why would a team move from Cartesia to Bland Speech?
A team moves from Cartesia to Bland Speech when it wants one standard per-character rate with no plan to size, and a voice that sounds like a person on a phone line.
As of November 2026, Cartesia's credits come with several rules. Overages are opt-in, and with them off, "new requests are blocked once your credits run out". Break tags in a request cost a credit each, and adding an accent to a voice costs 225 credits per added accent. Credits roll over up to twice the monthly allotment. Bland has no allowance to track: it charges $0.04 per 1,000 characters as audio is delivered, and an account with no balance returns an insufficient credits error, which is the same stopping point reached by a different route.
The reasons to stay are real too. Cartesia says Sonic delivers first audio in under 90ms, against Bland's published 199ms. Cartesia's Sonic 3.6 covers 44 languages. Inside a full allowance, Cartesia's larger plans also work out below Bland's rate.
What maps to what when moving from Cartesia to Bland Speech?
Cartesia's WebSocket with contexts maps to Bland's WSS /v2/tts/ws with a context_id per turn, its /tts/bytes maps to POST /v2/tts, and its pcm_mulaw maps to mulaw at 8,000 Hz.
What these rows mean
- Streaming text in: Bland describes a Cartesia move as mapping contexts and continuations to turns, preemption, and cancellation. On Bland, one
initmessage opens the session, text arrives asspeakdeltas under acontext_id, and aspeakunder a newcontext_idcuts off the turn that is playing. - Phone audio: Bland's encodings are
pcm_s16leandmulaw, with mu-law fixed at 8,000 Hz. - Delivery: Bland's
/v2/ttsbody takesexpressivenessandstability, and pacing is written into the text. Cartesia's speed, volume, and emotion controls come throughgeneration_configor SSML. - Price: Cartesia's plans are Free at 20K credits, Pro at $5 with 100K, Startup at $49 with 1.25M, and Scale at $299 with 8M.
Delivery needs the most thought. Cartesia sets pace and loudness with numbers, and Bland Speech takes pacing from the text, so a team moves pacing into the text with pause markers written <|0.8|> (a length from 0.1 to 10.0 seconds) and punctuation, and moves emphasis into capitals or tags. Bland publishes a verbatim tag set, such as [say warmly], [say quietly], and [laughs], placed before the words they shape, and says invented variants are ignored or garbled. The two controls, expressiveness and stability, each run from 0.0 to 1.0.
A team that relies on pronunciation dictionaries changes the text itself, spelling a name or number the way it should be said.
What are the steps to switch from Cartesia to Bland Speech?
A switch takes six steps: inventory endpoints, build a voice table, port the context model, request phone audio directly, rewrite delivery, and run both providers on a share of traffic.
- A team lists which Cartesia endpoint each flow calls (bytes, SSE, or WebSocket), because Cartesia's WebSocket is the endpoint that takes streamed text, and it decides where Bland's WebSocket goes.
- A team builds a voice table with each Cartesia voice id in one column and a Bland voice UUID beside it. A GET to
/v1/voiceslists every voice, and Karen, Matthew, and River are open to every organization. - A team ports the context model. A Cartesia context becomes one Bland turn, a follow-on piece of text for the same context becomes another
speakunder the samecontext_id, and an interruption becomes either a newcontext_idor acancelmessage:
{ "type": "init", "voice": "29158307-9893-4149-8a75-bc9ce313d64e", "audio": { "encoding": "mulaw", "sample_rate": 8000 } }
{ "type": "speak", "context_id": "turn-42", "text": "Hello" }
{ "type": "speak", "context_id": "turn-42", "text": ", how can I help?" }
{ "type": "end_of_turn", "context_id": "turn-42" }
{ "type": "cancel", "context_id": "turn-42" }
- A team that asks Cartesia for
pcm_mulawat 8,000 Hz asks Bland for{"audio": {"encoding": "mulaw", "sample_rate": 8000}}, and a team onpcm_s16lekeeps that encoding and picks one of Bland's five rates. A team on 22,050 Hz picks 24,000 or 16,000. - A team rewrites speed, volume, and emotion settings as pause markers, punctuation, tags, and the two controls, and renders each line in Speech Studio, which uses the same models as the API, before putting it in code.
- A team routes a small share of calls to Bland, compares recordings against Cartesia output for the same lines, and moves the rest once the numbers hold.
Bland returns audio as binary frames until it sends utterance_end for the turn, and the client discards audio it hadn't played yet when a turn is cut off. Text goes out as a language model writes it, with its leading spaces and punctuation intact, because Bland picks the synthesis boundaries itself.
Will callers hear a difference after moving from Cartesia to Bland Speech?
Bland Speech v3 ranked first of 13 AI voices in Design Arena's Audio Realism Bench results of 4 August 2026, at 1365, with the human recording at 1500, while Cartesia's Sonic 3.5 ranked 6th at 1096.
- #1BlandBland Speech v31365
- 2ndMicrosoftMAI-Voice-21224
- 3rdxAIGrok TTS1156
- 4thOpenAIGPT Realtime 21119
- 5thGoogleGemini 2.5 Pro TTS1109
- 6thCartesiaSonic 3.51096
Ranks are among 13 AI voices, and Cartesia's own voice is on the list. Real human recordings scored 1500, at the right edge of this scale. Bars start at 600, not zero. Source: Design Arena's published results.
Cartesia's Sonic 3.6 was released on 27 August 2026, after those results, and wasn't scored in them. The benchmark has native speakers pick, blind, which of two clips of the same text sounds more like a real person, and it has a Phone Agents category. A score across hundreds of prompts can't say how a given team's lines sound, so the check that settles it is a call script's most repeated lines rendered by Sonic 3.6 and by Bland, played to a listener who can't see the labels.
What does the move cost, and what can't come along?
Bland Speech costs $0.04 per 1,000 characters ($40 per million), pay as you go with no monthly plan, and 50,000 characters are free once for a new account. Cartesia's plans make the comparison depend on volume.
At 100,000 characters a month, Cartesia's Pro plan is $5 and Bland comes to $4. At 1 million characters, Cartesia's Startup plan is $49 for an allowance of 1.25 million, against $40 on Bland. At 8 million characters, Cartesia's Scale plan is $299, which works out to $37.38 per million, against $320 on Bland, so a team that fills a Scale allowance month after month pays less on Cartesia. Cartesia's overage rates sit at $65 per million credits on Pro, $45 on Startup, and $38 on Scale. Bland adds a $0.001 minimum per request and bills for what is delivered, so a synthesis that fails before any bytes are written isn't charged. Enterprise pricing is custom, set with Bland's sales team.
Cartesia voices don't come along. Bland says voice IDs are provider-specific and can't be reused, so each voice is cloned again on Bland from its original recording. An instant clone needs one clean sample of about 10 seconds, a professional clone is fine-tuned on 30 minutes or more of audio and goes through Bland's consent and rights review, every clone is private to the customer's organization, and the customer confirms the voice is theirs to use.
Cartesia also states that Sonic-3.6 can be deployed on-prem inside a data center, including air-gapped environments, in a customer's own VPC, or through OEM licensing, under enterprise contracts. Bland's Enterprise customers get dedicated orchestration and dedicated GPUs run by Bland. A team that runs Cartesia's Line SDK agents has a separate date to plan around, since Cartesia stops hosting Line SDK agents on 1 January 2027.
A Cartesia review and a list of Cartesia alternatives cover the provider in more depth, and Bland Speech vs Cartesia voice cloning covers clones. Moves from other providers have their own text to speech migration guides, the text to speech comparison hub holds every comparison, the text to speech API page lists each endpoint, and Bland Speech has the studio.
Frequently asked questions
How do Cartesia contexts map to Bland Speech?
A Cartesia context maps to one Bland turn, identified by a context_id that a team picks. Text deltas for the turn go out as speak messages under that id, end_of_turn closes it, and a speak under a new id or a cancel message handles an interruption.
Does Cartesia's 8 kHz mu-law output carry over?
Yes: pcm_mulaw at 8,000 Hz becomes mulaw at 8,000 Hz on Bland's /v2/tts and /v2/tts/ws, and Bland also returns pcm_s16le at 8,000, 16,000, 24,000, 44,100, and 48,000 Hz.
Can a Cartesia voice move to Bland Speech?
No. A voice is tied to the provider that made it, so a team clones it again on Bland from its original recording. An instant clone needs one clean sample of about 10 seconds, and a professional clone needs 30 minutes or more and passes Bland's consent and rights review.
Is Bland Speech cheaper than Cartesia?
The answer depends on volume. At 100,000 characters a month Bland comes to $4 against $5 on Cartesia's Pro plan, and at 1 million it is $40 against $49 on Startup, but Cartesia's Scale plan works out to $37.38 per million inside its 8 million allowance, below Bland's $40.
Is Bland Speech as fast as Cartesia's Sonic?
Cartesia says Sonic delivers first audio in under 90ms, and Bland publishes 199ms to first audio on a streamed request. Those are two companies' own figures, so a team that cares about the difference tests both on its own network.
What happens to Cartesia's speed, volume, and emotion settings?
They move into the text and two controls. Pacing goes into pause markers and punctuation, emphasis moves into capitals and tags, and two controls, expressiveness and stability, replace the numeric settings.
Do Pipecat and LiveKit agents need code changes to move?
A small change in each. The Pipecat team maintains services for both Cartesia and Bland, and Bland's installs with uv add "pipecat-ai[bland]". LiveKit Agents has a Bland plugin for Python, installed with uv add "livekit-agents[bland]~=1.8", so the swap is an install and a class change.
