Speechify's new Simba 3.2 text-to-speech model has claimed the top spot on Artificial Analysis's independent voice leaderboard, outperforming rivals like OpenAI and ElevenLabs in blind listener tests. The company also introduced a developer platform and AI voice-agent service.
TL;DR
- Speechify's Simba 3.2 AI voice model leads Artificial Analysis's text-to-speech leaderboard with an Elo rating of around 1,234.
- The model offers real-time performance with a time to first audio of less than 100 milliseconds and supports SSML controls.
- Priced at $10 per one million characters (or $6 at Scale tier), Simba 3.2 is significantly cheaper than competitors like ElevenLabs and Cartesia.
What happened
Speechify has launched Simba 3.2, a real-time text-to-speech model that has secured the first position on Artificial Analysis's Text-to-Speech Arena leaderboard. The model outperformed competitors from Google, OpenAI, ElevenLabs, Cartesia, and others in blind listener comparisons. Additionally, Speechify introduced a developer platform and AI voice-agent service powered by Simba 3.2.
Simba 3.2 is designed for real-time applications, delivering audio in under 100 milliseconds. It supports SSML controls for pitch, speaking speed, pauses, and emphasis, as well as emotional expression and word-level timing. The model currently supports English, with non-English requests directed to Speechify's Simba Multilingual model.
Speechify's API pricing for Simba 3.2 starts at $10 per one million characters, dropping to $6 per one million characters for Scale tier customers. This pricing makes Simba 3.2 more than 15 times cheaper than ElevenLabs and around six times cheaper than Cartesia, according to the company.
Why it matters
For developers and startups, Simba 3.2 offers a high-quality, low-latency text-to-speech solution at a competitive price. This could be particularly beneficial for applications requiring real-time voice interaction, such as voice agents, customer-service systems, and accessibility tools.
Investors may see Speechify's advancements in AI voice technology as a strategic move to capture market share in the growing text-to-speech industry. The company's focus on efficiency and lower pricing could attract businesses processing large volumes of voice-related content.
However, the long-term position of Simba 3.2 will depend on how it performs as more users test it and rival providers release updated models. The current rankings are based on initial evaluations, and the competitive landscape may evolve.
Key facts
- Simba 3.2 holds first place on Artificial Analysis's Text-to-Speech Arena leaderboard with an Elo rating of around 1,234.
- The model is third on Voice Arena's US English leaderboard, with a confidence range that overlaps with the second-ranked model.
- Simba 3.2 delivers audio in under 100 milliseconds, making it suitable for real-time applications.
- Pricing starts at $10 per one million characters, with a Scale tier rate of $6 per one million characters.
- Speechify's consumer applications have served tens of millions of users, providing feedback and testing data for Simba 3.2.
- The developer platform supports streaming speech, voice cloning, emotional controls, and SSML support.
- Simba 3.2's API accepts up to 2,000 characters through its standard speech endpoint and up to 20,000 characters per streaming request.
Context
The AI voice market is rapidly growing, with applications ranging from virtual assistants to accessibility tools. High-quality, low-latency text-to-speech models are in demand for real-time interactions. Speechify's Simba 3.2 aims to address these needs with competitive performance and pricing.
Artificial Analysis's leaderboard uses blind listener voting to generate Elo scores for text-to-speech models. This method provides an independent assessment of model quality, helping developers and businesses make informed decisions.
Speechify's focus on efficiency and lower pricing could disrupt the market, making advanced text-to-speech technology more accessible to a broader range of applications and businesses.
