On November 13 Artificial Analysis said Speechify's Simba 3.5 text to speech model is #1 in its Controlled Voice Arena, statistically level with HeyGen Voice. Every model on this board reads with the same 8 cloned voices (4 US, 4 UK) and users vote blind, so it compares models rather than showcase voices. The leaderboard lists Simba 3.5 at 1192±17 Elo from 1,783 samples, HeyGen Voice at 1190±13 and Qwen-Audio-3.1-TTS-Plus third at 1176; Simba 3.5 costs $6.58 per 1M characters on the API versus $30.00 for HeyGen Voice.
Key Takeaways
- ✓The Controlled Voice Arena has every model use the same 8 cloned voices (4 US, 4 UK) with blind user votes, removing the bias of picking a flattering demo voice
- ✓Simba 3.5 scores 1192±17 Elo (95% interval 1175-1209, 1,783 samples) with a rank range of 1-3; HeyGen Voice scores 1190±13 with a range of 1-2, so the two are statistically level
- ✓Third to fifth: Qwen-Audio-3.1-TTS-Plus 1176, HeyGen Voice Turbo 1174, Qwen-Audio-3.1-TTS-Flash 1164; Eleven v4 Turbo (1156) and Eleven v4 (1155) are sixth and seventh
- ✓Price per 1M characters on each creator's API: Simba 3.5 $6.58, HeyGen Voice $30.00, HeyGen Voice Turbo $4.00, Eleven v4 $80.00
- ✓Speechify's previous Simba 3.2 sits at 1082 Elo in 14th on the same board; AA's post also puts Simba 3.5 fifth on the Provider Voice board, which uses each vendor's own voices
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background
Text to speech leaderboards often let each vendor enter its most flattering demo voice, so votes reward the voice more than the model. Developers who build audiobooks, dubbing or voice agents on cloned voices care about which model reads the same voice most naturally. Artificial Analysis built the Controlled Voice Arena for that question.
Architecture highlights
Every model on the board reads the same content with the same 8 cloned voices (4 US, 4 UK). Users compare pairs blind and vote, and results become Elo scores with 95% confidence intervals and rank ranges, anchored to OpenAudio S1 at 1000. Speechify itself notes that this setup measures the model, not the demo voice.
Benchmarks and data
Per AA's post and leaderboard: Simba 3.5 is first at 1192±17 (1,783 samples), statistically level with HeyGen Voice at 1190±13; Qwen-Audio-3.1-TTS-Plus (1176) and HeyGen Voice Turbo (1174) follow, and Eleven v4 scores 1155. Speechify's previous Simba 3.2 scores 1082. AA's post says Simba 3.5 ranks fifth on the Provider Voice board.
Developer quick start
If your product runs on cloned voices, shortlist Simba 3.5 and HeyGen Voice: quality is close, but Simba 3.5 costs $6.58 per 1M characters against $30.00 for HeyGen Voice; on a tighter budget, look at HeyGen Voice Turbo ($4.00, 1174 Elo). The board only tests US and UK English voices, so re-test other languages on your own samples.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.