Simba 3.5 Ranks #2 of 51 AI Voice Models at $6 per 1M Characters. Eleven v4 Costs $80. On November 21, Simba Rises 67%.
Speechify’s Simba 3.5 is the #2 voice on Artificial Analysis’s blind Controlled Voice arena, 2 Elo points behind HeyGen Voice, and bills $6 to $10 per million characters. Eleven v4 sits 34 points lower at #6 and lists at $80. On November 21, Simba’s launch rate ends and its price rises 50% to 67%.
What changes on November 21
Speechify launched Simba 3.5 on October 21 as its recommended text-to-speech model: one model id, simba-3.5, for 48 languages, up from 6 in Simba 3.0. For its first month it bills at the rates of the older Simba 3.0 and 3.2. That month ends on November 21.
| Plan (monthly fee) | Simba 3.5 until Nov 21 | Simba 3.5 from Nov 21 | Rise |
|---|---|---|---|
| Starter ($10) | $10 per 1M characters | $15 per 1M characters | +50% |
| Pro ($99) | $8 per 1M characters | $12 per 1M characters | +50% |
| Scale ($499) | $6 per 1M characters | $10 per 1M characters | +67% |
| Simba 3.0 and 3.2, any plan | $10 / $8 / $6 | $10 / $8 / $6 | none |
Speechify’s pricing page and Simba 3.5 launch post, opened November 16, 2026. The rise column is our arithmetic.
Each paid plan turns its fee into a dollar balance, so the rise shows up as fewer characters. Starter’s balance buys 2M characters of Simba 3.5 until November 21 and 1.3M after. Pro drops from 13.6M to 9M, Scale from 99.8M to 59.8M, and the free tier from 500K a month to 333K.
Two more dates sit next to it. New subscriptions have started on these plans since November 12, and existing subscribers move to them at their first renewal on or after November 22. And from November 21, the retired Simba 1.6 id simba-multilingual is served and billed as Simba 3.5, so old code that still names it starts paying the new model’s rate without changing a line.
The leaderboard, with prices
Artificial Analysis runs two blind voice arenas. On the Controlled Voice board, every model reads in the same 8 voices, 4 US and 4 UK English, so voters judge the model rather than the voice actor. On November 16 it ranked 51 models. Here is the top 9, plus 3 familiar names further down:
| Rank | Model | Elo | Rank range | Price per 1M characters |
|---|---|---|---|---|
| 1 | HeyGen Voice | 1191 | 1-2 | $30.00 |
| 2 | Simba 3.5 | 1189 | 1-4 | $6.58 |
| 3 | HeyGen Voice Turbo | 1177 | 1-5 | $4.00 |
| 4 | Qwen-Audio-3.1-TTS-Plus | 1174 | 3-5 | $19.31 |
| 5 | Qwen-Audio-3.1-TTS-Flash | 1167 | 3-7 | $19.31 |
| 6 | Eleven v4 | 1155 | 5-8 | $80.00 |
| 7 | Eleven v4 Turbo | 1155 | 6-9 | $40.00 |
| 8 | Cartesia Sonic 3.6 | 1148 | 6-9 | $49.00 |
| 9 | Inworld Realtime TTS-2 | 1143 | 7-9 | $20.83 |
| 14 | Simba 3.2 | 1081 | 13-17 | $6.58 |
| 16 | Eleven v3 | 1074 | 14-21 | $100.00 |
| 24 | Gemini 3.8 Flash TTS | 1052 | 21-26 | $16.49 |
Artificial Analysis, Controlled Voice Arena leaderboard, opened November 16, 2026. Prices are the board’s; Speechify’s own rate card for Simba 3.5 is $6 to $10 until November 21 and $10 to $15 after.
The top 3 are a statistical tie. Simba 3.5’s 95% interval is ±17 points, and its rank range runs from 1 to 4. Speechify’s own site still says #1: that was true on November 13, when it read 1,192 with a range of 1 to 3. Three days later HeyGen Voice is 2 points ahead. Neither reading is a quality gap.
The price gap is not a tie. Eleven v4 lists at $80 per 1M characters on the board, more than 13x Simba 3.5’s $6 Scale rate, and scores 34 points lower. Even after November 21, Simba 3.5 at $10 on Scale is one eighth of Eleven v4’s price.
HeyGen Voice Turbo is the cheapest model in the top 3. At $4.00 it costs less than Simba 3.5 at any rate, and it sits 12 points behind with an overlapping range.
On this board the 3 best-ranked voices cost $30, $6.58 and $4.00 per 1M characters. The two ElevenLabs models right below them cost $80 and $40.
Where ElevenLabs still wins
The other board tells a different story. On Artificial Analysis’s provider-voice leaderboard, each lab speaks in its own best voices, which is what you get when you pick a voice from a catalog. There, Eleven v4 and Eleven v4 Turbo tie for #1 at 1340, Qwen-Audio 3.1 TTS Plus is #3 at 1298, Cartesia Sonic 3.6 is #4 at 1295, and Simba 3.5 is #5 at 1291, 49 points behind.
Read the two boards together: Simba 3.5 the model is as good as anything when the voice is held constant, and ElevenLabs’ stock voices are still the ones listeners like most. If you use catalog voices for a brand narrator, test both. If you bring the same voice to every vendor, the price is the main difference.
What Speechify says it measured
Speechify’s own numbers, for English, as of October 2026:
| Claim | Figure |
|---|---|
| Blind preference over Simba 3.2 | 63% of comparisons (95% CI 60-67%) |
| Speech throughput per serving instance | 1.8x Simba 3.2 |
| Time to first audio, 95% of streaming requests | 72 ms, under production load |
| Preference over Simba 3.0 Multilingual, 6 shared locales | 62% of comparisons |
| Languages | 48 (Estonian and Cantonese not included) |
The limits are as useful as the wins. Simba 3.5 applies 3 SSML tags, pauses with break, speaking rate with prosody rate and spoken substitutions with sub alias; pitch, volume, emphasis and emotion tags are accepted and ignored. A streaming request takes up to 20,000 spoken characters. And a request without a model field still gets Simba 3.0, so the upgrade only happens if you name it.
Switching takes 1 field
Speechify’s launch post shows the whole change as one request. This one streams a German line from a German catalog voice into an MP3:
curl https://api.speechify.ai/v1/audio/stream \
-H "Authorization: Bearer $SPEECHIFY_API_KEY" \
-H "Content-Type: application/json" \
-H "Accept: audio/mpeg" \
-d '{
"input": "Ein Modell, achtundvierzig Sprachen.",
"voice_id": "amelie",
"model": "simba-3.5",
"language": "de-DE"
}' \
--output speech.mp3The same model value works on the batch endpoint, the timestamps endpoint and the WebSocket. Voice IDs, output formats and speech marks carry over from Simba 3.2, and no API version change is needed. Set language on every request whose text isn’t in the voice’s own language; without it, the voice’s locale decides. Both ids answer on the same endpoints, so you can move one call at a time and compare the output before November 21.
What a voiceover costs next to the video
Voice is the cheap part of an AI video. A 1,000-character script costs $0.006 to $0.01 on Simba 3.5 until November 21, $0.01 to $0.015 after, and $0.08 at Eleven v4’s board price. Our arithmetic, per finished voiceover:
| Model and rate | 1,000 characters | 7 scripts a week | 100K characters |
|---|---|---|---|
| HeyGen Voice Turbo, $4.00 | $0.004 | $0.028 | $0.40 |
| Simba 3.5 Scale until Nov 21, $6 | $0.006 | $0.042 | $0.60 |
| Simba 3.5 Starter from Nov 21, $15 | $0.015 | $0.105 | $1.50 |
| HeyGen Voice, $30 | $0.03 | $0.21 | $3.00 |
| Eleven v4, $80 | $0.08 | $0.56 | $8.00 |
On Simba 3.5, a week of 7 scripts costs $0.042 to $0.105 in voice, less than 1 second of Wan 3.0, the #1 model on Artificial Analysis’s text-to-video board, at $12.00 a minute. The voice price only matters at volume: an app that reads 100M characters a month pays $600 on Simba 3.5 Scale now and $1,000 after November 21.
What this means for your content
If you’re on Speechify, lock in your tests this week. Anything you synthesize on Simba 3.5 before November 21 bills at the old rate. Batch-render the voiceovers you already have scripts for.
If you still call simba-multilingual, check your bill. From November 21 that id runs on Simba 3.5 at Simba 3.5’s price.
For anything beyond English and Chinese, add the voice in the edit. Native audio in video models is limited: Kling 3.0’s sound option speaks English and Chinese only. Generate the clip without sound and lay a TTS voiceover on top. Simba 3.5 covers 48 languages, from Russian and German to Hindi and Japanese, under one model id.
Pick by the board that matches how you work. Same narrator across vendors: compare on the Controlled Voice board, where Simba 3.5 and HeyGen Voice lead. Stock catalog voices: the provider-voice board, where ElevenLabs leads.
For a full pipeline, our week of Reels from 1 product photo walks through scenes, clips and captions with the cost of each step, and the media API docs show how to turn a model’s sound off in the request.
Try it on MegaBrain
MegaBrain doesn’t sell Speechify’s voices. It does serve the video side of a narrated clip: Kling 3.0 Standard costs $0.0966 per second of video without sound and $0.1449 with sound, so a 5-second silent clip costs $0.483. One API key, one pay-as-you-go balance, and the same key runs image and text models.
Sources
Pages opened November 16, 2026. Every number above comes from these pages or MegaBrain’s model page, or from our arithmetic on them where we say so.
- Artificial Analysis — Controlled Voice Arena leaderboard (Elo, 95% intervals, rank ranges, samples, price per 1M characters)
- Artificial Analysis — Text to Video leaderboard (Wan 3.0 price per minute at benchmark settings)
- Artificial Analysis — Text to Speech leaderboard, provider voices
- SpeechifyAI — Simba 3.5 launch post (October 21, 2026, updated November 13: languages, prices, measured results, SSML)
- SpeechifyAI — API pricing (plans, included characters before and after November 21, plan dates)
- SpeechifyAI — Text-to-Speech API page (endpoints, formats, 20,000-character stream limit)
- SpeechifyAI docs — Models (default model, retired Simba 1.6 ids, 48 language codes)
Full disclosure: MegaBrain sells pay-as-you-go API access to image, video and text models, including Kling 3.0. MegaBrain is not affiliated with Speechify, HeyGen, ElevenLabs, Alibaba, Cartesia, Inworld, Google, Kuaishou or Artificial Analysis.
MegaBrain for Mac · an AI coworker
Give MegaBrain the task, not the steps.
It works on your files, writes documents, spreadsheets and decks, connects to your team’s tools and asks before anything risky. Free to start.
Newsletter
Stay in the loop
Get our latest AI research and guides. No spam, unsubscribe anytime.