Home
ElevenLabs Sets New Speech-to-Text Benchmark; Google Gemini Follows with Broad Capabilities
Artificial Analysis has released the latest version of its speech-to-text benchmark, AA-WER v2.0. The findings highlight ElevenLabs and Google as clear leaders in audio transcription performance.

When measured by the core word error rate (WER), ElevenLabs' Scribe v2 achieved the top spot with an impressively low 2.3% error rate. Close behind was Google's Gemini3Pro at 2.9%. It's worth noting that Google did not fine-tune Gemini for transcription; this result stems purely from its robust multimodal general capabilities.
Other leading models showed the following results:
Mistral Voxtral Small: Took third place with a 3.0% error rate.
Google Gemini3Flash: Delivered a solid performance with a 3.1% error rate.
OpenAI Whisper Large v3: The most widely-used open-source model placed in the middle of the pack with a 4.2% error rate.
Lowest performers: Alibaba's Qwen3ASR Flash (5.9%), Amazon's Nova2Omni (6.0%), and Rev AI (6.1%) rounded out the bottom of the rankings.

In the dedicated AA-AgentTalk benchmark for voice assistant commands, the leaderboard remained consistent. ElevenLabs' Scribe v2 and Google's Gemini3Pro maintained their lead with error rates of 1.6% and 1.7% respectively, proving highly reliable for short, direct voice interactions.
Related article
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
Related Special Topic Recommendations
Comments (2)
0/500
ElevenLabs is really pushing the boundaries, but seeing Google jump in with Gemini makes the competition fierce. I'm curious how this impacts small devs trying to get affordable transcription tools. The WER numbers are impressive, yet real-world noise is still a pain point for me.
Artificial Analysis has released the latest version of its speech-to-text benchmark, AA-WER v2.0. The findings highlight ElevenLabs and Google as clear leaders in audio transcription performance.

When measured by the core word error rate (WER), ElevenLabs' Scribe v2 achieved the top spot with an impressively low 2.3% error rate. Close behind was Google's Gemini3Pro at 2.9%. It's worth noting that Google did not fine-tune Gemini for transcription; this result stems purely from its robust multimodal general capabilities.
Other leading models showed the following results:
Mistral Voxtral Small: Took third place with a 3.0% error rate.
Google Gemini3Flash: Delivered a solid performance with a 3.1% error rate.
OpenAI Whisper Large v3: The most widely-used open-source model placed in the middle of the pack with a 4.2% error rate.
Lowest performers: Alibaba's Qwen3ASR Flash (5.9%), Amazon's Nova2Omni (6.0%), and Rev AI (6.1%) rounded out the bottom of the rankings.

In the dedicated AA-AgentTalk benchmark for voice assistant commands, the leaderboard remained consistent. ElevenLabs' Scribe v2 and Google's Gemini3Pro maintained their lead with error rates of 1.6% and 1.7% respectively, proving highly reliable for short, direct voice interactions.
Stripe to Acquire OpenRouter as AI Model Routing Gains Momentum
Stripe has agreed to acquire OpenRouter, an AI model-routing platform that gives developers access to hundreds of models through a single interface. The deal adds model selection and routing to Stripe’s existing work around AI usage and token-based b
Former OpenAI Researcher Barrett Zoph Returns to Google to Support Gemini Development
Barret Zoph, who previously abruptly left and joined OpenAI due to a disagreement with Thinking Machines Lab CEO Miriam Mularati, has once again changed employers. Alphabet's Google announced that Zoph will return as Vice President of Research, bring
How Recursive Intelligence Raised $335M at $4B Valuation in 4 Months
The co-founders of startup Ricursive Intelligence seemed destined to be co-founders.Anna Goldie, CEO, and Azalia Mirhoseini, CTO, are so well-known in the AI community that they were among those AI engineers who “got those weird emails from Zuckerber
ElevenLabs is really pushing the boundaries, but seeing Google jump in with Gemini makes the competition fierce. I'm curious how this impacts small devs trying to get affordable transcription tools. The WER numbers are impressive, yet real-world noise is still a pain point for me.











