Cohere Challenges NVIDIA's Dominance with Open-Source Lightweight Speech Model Transcribe

Cohere, a long-time player in the enterprise AI space, officially launched an open-source speech recognition model called Cohere Transcribe on March 26, 2026.
The model, with 2 billion parameters, is built for edge devices to overcome the latency issues caused by larger speech models. By open-sourcing it under the Apache 2.0 license, Cohere follows Meta's lead, tapping into the developer community to rapidly improve the ecosystem and eventually drive commercial returns.
Edge Performance Powerhouse: 14 Languages Supported, Outperforming Top Competitors
Cohere Transcribe was trained on 14 languages, including Chinese, Japanese, French, and Hebrew. According to the latest Hugging Face open ASR leaderboard, the model has already surpassed rivals like ElevenLabs Scribe and Alibaba's Qwen3.
With fewer parameters, it can run directly on smartphones, PCs, or industrial gateways without frequent cloud calls. This cuts data transmission latency significantly and offers stronger privacy protection for sensitive sectors like banking, sales, and healthcare.
From Text to Speech: A Strategic Move to Redefine Agent Interaction
Though Cohere has focused on text generation, this foray into speech recognition is seen as a critical step toward building a full-fledged AI agent. The company says Cohere Transcribe will soon be integrated into its AI agent orchestration platform, North.
Analysts note that as voice interfaces like Siri become central to AI trends, speech capabilities are essential for agents to perceive the world. Cohere is taking on IBM, Alibaba, and Zoom (which launched AI Companion 3.0) with a "small but mighty" open-source strategy targeting edge computing and real-time speech translation.
Related article
Claude Opus 5.5 Enters Internal Testing, Leaps Two Tiers to Challenge GPT-6 at a Fraction of the Cost
While anticipation for Opus 5.2 remains high, recent developments indicate a strategic leap to version 5.5. Early today, developer Lyra shared insider information revealing that Claude Opus 5.5, internally codenamed "claude-wafer-eap," is currently u
OpenAI Unveils GPT-6-Powered Intelligent UI, Redefining ChatGPT’s Interactive Visual Experience
OpenAI is set to introduce a new "Intelligent UI" this Wednesday alongside its latest large language model, GPT-6, delivering a major leap forward in ChatGPT’s visual and interactive capabilities. This update moves beyond traditional text-only respon
PHP Armageddon: OpenAI, Google’s Wiz uncover ancient holes, sparking internet panic
The U.S. National Security Agency issued a critical alert on Wednesday regarding two severe vulnerabilities in PHP, for which patches have already been released by the PHP Foundation.As a server-side scripting language, PHP powers approximately 70% o
Related Special Topic Recommendations
Comments (0)
0/500

Cohere, a long-time player in the enterprise AI space, officially launched an open-source speech recognition model called
The model, with 2 billion parameters, is built for edge devices to overcome the latency issues caused by larger speech models. By open-sourcing it under the Apache 2.0 license, Cohere follows Meta's lead, tapping into the developer community to rapidly improve the ecosystem and eventually drive commercial returns.
Edge Performance Powerhouse: 14 Languages Supported, Outperforming Top Competitors
With fewer parameters, it can run directly on smartphones, PCs, or industrial gateways without frequent cloud calls. This cuts data transmission latency significantly and offers stronger privacy protection for sensitive sectors like banking, sales, and healthcare.
From Text to Speech: A Strategic Move to Redefine Agent Interaction
Though Cohere has focused on text generation, this foray into speech recognition is seen as a critical step toward building a full-fledged AI agent. The company says
Analysts note that as voice interfaces like Siri become central to AI trends, speech capabilities are essential for agents to perceive the world. Cohere is taking on IBM, Alibaba, and Zoom (which launched AI Companion 3.0) with a "small but mighty" open-source strategy targeting edge computing and real-time speech translation.
Claude Opus 5.5 Enters Internal Testing, Leaps Two Tiers to Challenge GPT-6 at a Fraction of the Cost
While anticipation for Opus 5.2 remains high, recent developments indicate a strategic leap to version 5.5. Early today, developer Lyra shared insider information revealing that Claude Opus 5.5, internally codenamed "claude-wafer-eap," is currently u
OpenAI Unveils GPT-6-Powered Intelligent UI, Redefining ChatGPT’s Interactive Visual Experience
OpenAI is set to introduce a new "Intelligent UI" this Wednesday alongside its latest large language model, GPT-6, delivering a major leap forward in ChatGPT’s visual and interactive capabilities. This update moves beyond traditional text-only respon
PHP Armageddon: OpenAI, Google’s Wiz uncover ancient holes, sparking internet panic
The U.S. National Security Agency issued a critical alert on Wednesday regarding two severe vulnerabilities in PHP, for which patches have already been released by the PHP Foundation.As a server-side scripting language, PHP powers approximately 70% o





Home






