# Speechify Simba 3.5 API Pricing: What Changed for Simba 1.6 Calls

The launch rate ended on November 21, and two old model IDs can still return audio while drawing from your balance at the new rate.

By Paras Madan | 2026-11-22


SpeechifyAI started charging its regular Simba 3.5 rate on November 22: $15 per million characters on Starter, $12 on Pro and $10 on Scale. That matters even if your code never says `simba-3.5`. Since November 21, requests that still use the old Simba 1.6 IDs on compatible API versions have been served and billed as Simba 3.5.

## Speechify Simba 3.5 pricing after the launch offer

Speechify originally published a higher list price and then revised it before the launch offer ended. Use the [November 9 pricing change](https://docs.speechify.ai/build/changelog/2026/11/9) and the [current API pricing page](https://speechify.ai/pricing) rather than the numbers crossed out on the October launch page.

| Paid plan | Simba 3.5 through Nov 21 | Simba 3.5 from Nov 22 | Simba 3.0 or 3.2 |
| --- | ---: | ---: | ---: |
| Starter | $10 | $15 | $10 |
| Pro | $8 | $12 | $8 |
| Scale | $6 | $10 | $6 |

All rates are in US dollars per million billable characters. These are **rates at which usage draws down your balance**, not another fee added to every subscription. As a simple example, on Pro, two million characters of English narration draw $24 on Simba 3.5 or $16 on Simba 3.2. That is an $8 difference in usage value, not necessarily an $8 increase on your invoice if both fit inside the included balance.

Speechify also distinguishes the model rate from when upgraded plan allowances apply. Its pricing page says existing subscriptions move to the new plan terms at their first renewal on or after November 22. Check the workspace's renewal date and available balance rather than assuming both changes happen together. On the Free plan, the advertised 500,000 characters refer to the cheaper models; Speechify lists about 333,000 for Simba 3.5 and pauses requests once the balance runs out. These are API plans, separate from the Speechify Reader subscription.

## Why Simba 1.6 can keep returning audio

Speechify changed what the two old IDs mean, and what happens depends on the API version. The [retirement guide](https://docs.speechify.ai/build/migrating-off-simba-1-6) says that on version `2026-09-21` or later, `simba-english` and `simba-multilingual` return `400 model_retired`. On versions that still accept those names, both now produce Simba 3.5 audio, including English requests. The [November 10 correction](https://docs.speechify.ai/build/changelog/2026/11/10) supersedes the earlier plan to route English calls on the old English ID to Simba 3.2.

Pinning an older API version can explain why an old request still succeeds. It cannot bring back the original Simba 1.6 voice after the November 21 switch. Previously generated files are unaffected. Calls that omit the model still default to Simba 3.0, a separate behavior worth keeping in mind when you audit what you pay for.

## Test the voice before you change the string

Search request builders, worker configs and scheduled jobs for both aliases. Then take a few representative lines, use your existing voice ID and try them with the explicit replacement. For an English-only narration job, compare Simba 3.2 against 3.5 before paying the higher rate. For multilingual work, check your actual locales and set `language` explicitly. Speechify's guide says that when moving from the old multilingual model, omitting that field on explicit Simba 3.5 can use the voice's locale instead of the old model's text detection.

Include at least one script with a pause, a name and a line whose emphasis matters. Speechify documents differences in how its newer models handle some SSML emphasis and emotion tags, so a successful response alone will miss problems a listener hears. Compare the new output with a previously saved clip if you have one, then check the project's character usage and cost before migrating a large batch.

An old model ID that still returns audio can keep a job running while quietly changing what it sounds like and what it costs. Make the model choice explicit so the person reviewing the next voice update can see both decisions.

I'm Paras. I work across AI, engineering and distribution.

I help teams make API changes and migration guidance clear enough for developers to act on.

[Book a 30-minute call](https://cal.com/paras-madan/30min).
