The Pulse
SpaceXAI Releases Grok Voice Transcribe 2.0 at the Same Price
SpaceXAI says Grok Voice Transcribe 2.0 doubles the accuracy of its predecessor in real-world evaluations while keeping batch and streaming prices unchanged.
AI.info Team ·
SpaceXAI released Grok Voice Transcribe 2.0 on September 18, presenting it as an accuracy upgrade for speech-to-text workloads without raising the published price. The company says the model is twice as accurate as Grok Voice Transcribe 1.0 in its real-world evaluations.
Grok Voice Transcribe 2.0 is available for batch and streaming transcription through SpaceXAI's Speech-to-Text API. Batch transcription remains $0.10 per hour of audio, while streaming transcription costs $0.20 per hour. Diarization, timestamps and key terms are included at those rates, according to SpaceXAI.
Grok Voice Transcribe 2.0 Targets Noisy Audio
SpaceXAI built the model on the audio foundation model behind Grok Voice. The company says that system already supports tens of thousands of customer-support calls each day, transcribes millions of hours of video narration and powers voice agents in physical products, including the Grok assistant in Tesla vehicles.
The new model is trained on live, noisy and multilingual audio gathered across different environments. SpaceXAI specifically points to flaky phone lines, overlapping speakers, regional accents and spoken phone numbers or email addresses as conditions that expose weaknesses in systems tuned mainly for clean, single-speaker recordings.
On the public Artificial Analysis leaderboard, SpaceXAI says Grok Voice Transcribe 2.0 ranks first for accuracy among 32 streaming models. That result is the company's description of the leaderboard standing, rather than an independent test published with the launch.
Short Multilingual Phrases Show the Largest Gain
SpaceXAI reports improvements over version 1.0 across four internal evaluation sets: 8 kHz English telephony audio from customer-support calls, English conversations with Grok, spoken credentials such as account codes and email addresses, and short voice commands in 19 languages.
The largest reported improvement comes from short multilingual phrases, where the model must identify a language with little surrounding context. SpaceXAI says word error rate on that set falls from 20.6% with Grok Voice Transcribe 1.0 to 6.8% with version 2.0. Lower word error rates indicate fewer transcription mistakes.
The model can transcribe dozens of languages, detect the language automatically and follow switches between languages during a single recording. SpaceXAI describes multilingual accuracy as its largest improvement over the previous version.
Loom and Cursor Form SpaceXAI's Showcase Workflow
SpaceXAI uses a workflow involving Atlassian's Loom and the coding tool Cursor to show how better transcription can feed other software. A user can record an action plan in Loom, send the transcript into Cursor and use the resulting context to make code changes.
“We've always believed the best way to move work forward is to capture context once and let it flow everywhere. With Grok powering Loom's speech-to-text and Cursor turning that into code, we're closing the loop from context to code: record what you mean, and the work gets done. It's a glimpse of where AI-assisted development is headed.”
Sanchan Saxena, SVP of Teamwork Collection, Atlassian
The example reflects a broader product choice in the release: SpaceXAI is selling transcription as an input layer for other applications rather than only as a standalone transcript generator. Batch processing handles recorded files and URLs, while streaming supports live audio through the API.
Version 1.0 Will Be Deprecated During Migration
Existing Speech-to-Text API integrations receive the accuracy improvement without code changes. SpaceXAI says Grok Voice Transcribe 2.0 will soon become the default in the Speech-to-Text API, while Grok Voice Transcribe 1.0 will be deprecated in the coming weeks.
Developers who need to remain on the earlier model during the transition can pin grok-voice-transcribe-1.0.
SpaceXAI's release keeps the commercial decision straightforward: developers receive a new transcription model at the existing rates, while the technical change lies in accuracy across phone audio, spoken credentials, multilingual commands and other conditions that are difficult to handle reliably.