Best Open ASR Models of 2026: WER, Languages, Latency and Licenses Compared

The open speech recognition (ASR) market is no longer a monoculture of Whisper: over the past 12 months, many competitive models have emerged. The leaders of the Open ASR Leaderboard are separated by less than one percentage point in WER, so the decisive factors have become license, language coverage, streaming support, and cost per hour of audio. The article compares the main models in detail across these parameters and provides recommendations for selection.

Cohere IBM NVIDIA Alibaba/Qwen Mistral Kyutai Meta OpenAI

West Responds to China: Poolside Releases Most Powerful Open Model Laguna S 2.1

Startup Poolside has unveiled Laguna S 2.1, an open-source model for coding with 118 billion parameters, claiming it is the most powerful open model from the West. It is comparable in performance to models several times larger but lags behind China's Kimi K3. The release comes amid discussions in the US about a possible ban on using Chinese open-source models.

Poolside OpenAI DeepSeek NVIDIA Moonshot AI