SpaceXAI Releases Grok Voice Transcribe 2.0 at Its Existing API Prices

SpaceXAI has released Grok Voice Transcribe 2.0, promising twice the accuracy of its previous speech-to-text model at unchanged rates. The immediate offer is straightforward: developers can use it for recorded audio or real-time streams. The harder question is how broadly its performance claims will translate, since its headline comparison and several detailed results are supplied by the company itself.

The model is listed as available in SpaceXAI’s Speech-to-Text API. Its release notes say customers can select either version 1.0 or 2.0, with version 1.0 still the default. SpaceXAI says 2.0 will soon take that role and that it plans to deprecate 1.0 in the following weeks; customers wanting the older version during the transition can pin it.

For existing integrations, SpaceXAI says the new model’s accuracy improvement requires no code changes. Both batch processing and streaming remain supported, which puts the same release in two common workflows: transcribing completed recordings and producing text from live audio.

SpaceXAI positions 2.0 for conditions that frustrate transcription systems, including poor phone connections, competing speakers, accents, and spoken account details. It says the model automatically detects languages, follows a switch in language within one recording, and covers dozens of languages.

The company also says 2.0 leads 32 streaming models for accuracy on the public Artificial Analysis leaderboard and performs better than version 1.0 across four internal sets. Those include customer-support telephony, conversations with Grok, spoken credentials, and short multilingual commands. The public leaderboard claim offers one external reference point, but the release does not provide independent validation of the company’s broader internal comparisons.

Transcript controls included in the API

• Word-level timestamps and confidence scores, plus speaker labels through diarization.
• Independent transcription for up to eight audio channels.
• Key-term biasing, formatting for items such as numbers and email addresses, filler-word removal, and speaker-turn detection.

SpaceXAI lists batch transcription at $0.10 per audio hour and streaming at $0.20 per hour, with diarization, timestamps, and key terms included. Keeping that price schedule makes this a replacement decision rather than a new budget line for current users, provided its claimed gains apply to their audio.

SpaceXAI says Atlassian’s Loom now uses the model to transcribe every video after finding it more accurate than its previous solution. That is a named deployment, not a measure of how the model will perform across other recordings, languages, or production setups. For developers, the near-term decision is practical: test the new default candidate on the messy calls, commands, and recordings their products actually receive.