The company’s flagship model has performed strongly across Artificial Analysis and Voice Arena, while focusing on being competitively priced relative to several other highly rated options.Artificial Analysis and Voice Arena are two ranking platforms that developers may consult when comparing voice models, and recent results have placed the same model among the stronger performers. Both rely on blind human listening rather than self-reported numbers, and both, this month, placed the same model at or near the top of the field.Cliff Weitzman, the CEO and founder of SpeechifyAI, leads the research team behind that model, Simba 3.2, a streaming text-to-speech system built for real-time voice applications. In a recent post announcing the milestone, Weitzman credited the team behind the work, calling the release "one of their biggest milestones to date" and highlighting his brother and co-founder, Tyler Weitzman, who leads AI research, alongside the broader AI team.The original challenge behind voice AIVoice AI benchmarks have often been shaped by models developed in large research labs, some of which are priced primarily for enterprise use rather than smaller-scale developer applications. According to the company, SpeechifyAI made a different set of architectural decisions at the start of its research, approaching quality, latency, and cost as interconnected priorities and aiming to balance high-quality output with production-ready pricing.According to the company, those decisions were shaped, in part, by scale. Speechify says its consumer platform serves more than 60 million users and that insights from millions of A/B tests across that user base inform how its models are trained and evaluated. The company was recognized for that consumer platform with a 2025 Apple Design Award for inclusivity."SpeechifyAI is the AI voice provider your CFO will love," Rohan Pavuluri, Chief Business Officer at SpeechifyAI, shared. "We spent years improving how our models run cost efficiently because our consumer business required and we produced some of the best voices on the planet. That investment is how we can now put the top TTS model in the world on our developer platform at about as cheap as it comes. Many AI labs obsess over benchmarks and charge enterprise pricing. We focused on listeners and set prices for massive production use cases."A solution to the voice AI problemThe Artificial Analysis text-to-speech leaderboard runs a consistent, objective evaluation across commercially available voice models and is tracked widely across the voice AI industry. Simba 3.2 currently ranks ahead of several other prominent AI voice models.Voice Arena uses a blind-comparison format similar to Chatbot Arena. Native-speaking listeners hear two clips generated from the same text without being told which models produced them, then select the one they find more natural. The results are aggregated into an Elo-based rating. SpeechifyAI’s Simba 3.2 is among the models competing on these measures. As of July 2026, independent ratings published by Artificial Analysis placed Speechify’s flagship streaming text-to-speech model, SIMBA 3.2, at the top of the text-to-speech leaderboard alongside models from ElevenLabs, Cartesia, OpenAI, and Google DeepMind.For Tyler Weitzman, the result marks a long arc. "As an undergrad at Stanford, CS229 with Andrew Ng sparked my interest in ML. My course project was fine-tuning Tacotron 2," he posted on X. "My team's new model at Speechify just hit SOTA, five years later. It's pretty surreal looking back."The future of voice AIFor Cliff Weitzman, the future of voice AI is not something that will unfold over the next few years. It is something that has already started. The same model that now sits at the top of both leaderboards powers Speechify's consumer applications and is available to developers through SpeechifyAI Build on Speechify's Developer Platform, alongside the newly released SpeechifyAI Agents, all at speechify.ai. Additional languages and an even lower-cost model tier are on the roadmap."Simba 3.2 is our best model yet, now available onSpeechify.ai," Weitzman said. "It's built to power voice agents at scale and perfected from millions of A/B tests we run in our consumer platform. In TTS APIs, three things matter: cost, quality, and latency. Simba 3.2 has achieved SOTA on this trifecta. Beyond excited for you to experience it firsthand to power your experiences."VentureBeat newsroom and editorial staff were not involved in the creation of this content.
How Speechify’s Simba 3.2 gained ground across two well-known voice AI ranking platforms
The company’s flagship model has performed strongly across Artificial Analysis and Voice Arena, while focusing on being competitively priced relative to several other highly rated options.Artificial Analysis and Voice Arena are two ranking platforms that developers may consult when comparing voice models, and recent results have placed the same model among the stronger performers. Both rely on blind human listening rather than self-reported numbers, and both, this month, placed the same model at or near the top of the field.Cliff Weitzman, the CEO and founder of SpeechifyAI, leads the research team behind that model, Simba 3.2, a streaming text-to-speech system built for real-time voice applications. In a recent post announcing the milestone, Weitzman credited the team behind the work, calling the release "one of their biggest milestones to date" and highlighting his brother and co-founder, Tyler Weitzman, who leads AI research, alongside the broader AI team.The original challenge behind voice AIVoice AI benchmarks have often been shaped by models developed in large research labs, some of which are priced primarily for enterprise use rather than smaller-scale developer applications. According to the company, SpeechifyAI made a different set of architectural decisions at the start of its research, approaching quality, latency, and cost as interconnected priorities and aiming to balance high-quality output with production-ready pricing.According to the company, those decisions were shaped, in part, by scale. Speechify says its consumer platform serves more than 60 million users and that insights from millions of A/B tests across that user base inform how its models are trained and evaluated. The company was recognized for that consumer platform with a 2025 Apple Design Award for inclusivity."SpeechifyAI is the AI voice provider your CFO will love," Rohan Pavuluri, Chief Business Officer at SpeechifyAI, shared. "We spent years improving how our models run cost efficiently because our consumer business required and we produced some of the best voices on the planet. That investment is how we can now put the top TTS model in the world on our developer platform at about as cheap as it comes. Many AI labs obsess over benchmarks and charge enterprise pricing. We focused on listeners and set prices for massive production use cases."A solution to the voice AI problemThe Artificial Analysis text-to-speech leaderboard runs a consistent, objective evaluation across commercially available voice models and is tracked widely across the voice AI industry. Simba 3.2 currently ranks ahead of several other prominent AI voice models.Voice Arena uses a blind-comparison format similar to Chatbot Arena. Native-speaking listeners hear two clips generated from the same text without being told which models produced them, then select the one they find more natural. The results are aggregated into an Elo-based rating. SpeechifyAI’s Simba 3.2 is among the models competing on these measures. As of July 2026, independent ratings published by Artificial Analysis placed Speechify’s flagship streaming text-to-speech model, SIMBA 3.2, at the top of the text-to-speech leaderboard alongside models from ElevenLabs, Cartesia, OpenAI, and Google DeepMind.For Tyler Weitzman, the result marks a long arc. "As an undergrad at Stanford, CS229 with Andrew Ng sparked my interest in ML. My course project was fine-tuning Tacotron 2," he posted on X. "My team's new model at Speechify just hit SOTA, five years later. It's pretty surreal looking back."The future of voice AIFor Cliff Weitzman, the future of voice AI is not something that will unfold over the next few years. It is something that has already started. The same model that now sits at the top of both leaderboards powers Speechify's consumer applications and is available to developers through SpeechifyAI Build on Speechify's Developer Platform, alongside the newly released SpeechifyAI Agents, all at speechify.ai. Additional languages and an even lower-cost model tier are on the roadmap."Simba 3.2 is our best model yet, now available onSpeechify.ai," Weitzman said. "It's built to power voice agents at scale and perfected from millions of A/B tests we run in our consumer platform. In TTS APIs, three things matter: cost, quality, and latency. Simba 3.2 has achieved SOTA on this trifecta. Beyond excited for you to experience it firsthand to power your experiences."VentureBeat newsroom and editorial staff were not involved in the creation of this content.
