SpaceXAI launches Grok Voice Transcribe 2.0 with multilingual streaming accuracy boost

On September 18, 2026, SpaceXAI announced the general availability of Grok Voice Transcribe 2.0 through its speech‑recognition API, identified as grok-voice-transcribe-2.0. The company positions the new service as a direct upgrade to the original Grok Voice Transcribe, emphasizing a two‑fold improvement in transcription accuracy while keeping the per‑hour cost unchanged.

The announcement highlighted that the upgraded model delivers the same pricing model as its predecessor, meaning customers continue to pay $0.10 per hour for batch processing and $0.20 per hour for streaming, yet they receive a markedly higher level of precision.

According to SpaceXAI’s own benchmarking, the model now achieves a word‑error rate (WER) of 6.8 % on short multilingual sentences, down from 20.6 % recorded by version 1.0. This represents roughly a 67 % reduction in errors for that specific test set, which covers 19 languages and simulates real‑world, mixed‑language utterances.

The company also pointed out that the lower WER is consistent across a variety of acoustic conditions, reinforcing the claim that the new engine is robust enough for both quiet office recordings and noisy call‑center environments.

Benchmark performance and public ranking

The company cites the public Artificial Analysis leaderboard, where Grok Voice Transcribe 2.0 currently holds the top spot for precision among 32 streaming transcription models. SpaceXAI notes that the ranking is based on independent evaluations, although the exact methodology of the leaderboard is not disclosed in the press release.

In addition to the headline WER improvement, internal tests on four proprietary datasets—telephone calls, conversational speech, dictated identifiers, and short multilingual phrases—showed consistent error reductions across all domains. The most dramatic gains were observed in the short‑phrase set, where the error drop from 20.6 % to 6.8 % was recorded.

These internal results were corroborated by third‑party auditors who confirmed that the error reduction holds even when the audio contains overlapping speakers and background music, scenarios that typically challenge speech‑to‑text systems.

Technical capabilities and API features

Grok Voice Transcribe 2.0 supports both batch and streaming modes. In batch mode, users can submit audio files for offline processing, while streaming mode delivers real‑time transcription with word‑level timestamps, speaker diarisation, and up to eight audio channels. All these features are included in the base price of $0.10 per hour for batch and $0.20 per hour for streaming.

The model automatically detects the spoken language and can switch detection mid‑recording, a capability that is especially useful for multilingual meetings or customer‑support calls where agents and callers may alternate languages.

Beyond basic transcription, the API returns formatted text, accentuation of a curated list of one hundred specialized terms, and speaker separation without additional fees. This bundled offering contrasts with many competitors that charge extra for diarisation or timestamping.

Adoption and ecosystem integration

Atlassian has already integrated Grok Voice Transcribe 2.0 into its Loom video platform to generate captions for recorded meetings and tutorials. The service remains hosted by SpaceXAI, and the company has not released any open‑weight model for on‑premises deployment.

• $0.10 per hour for batch transcription
• $0.20 per hour for streaming transcription
• Automatic language detection for 19 languages
• Speaker diarisation and word‑level timestamps included

The pricing structure is straightforward: customers are billed per hour of audio processed, with no hidden fees for advanced features. This simplicity is intended to appeal to enterprises that need predictable cost models for large‑scale transcription workloads.

What this means for English‑speaking organisations

For organisations that operate primarily in English but interact with multilingual clients or partners, Grok Voice Transcribe 2.0 offers a single API that can handle mixed‑language content without the need to route audio to separate language‑specific services. The reduced error rate on short phrases improves the reliability of automated note‑taking, compliance logging, and real‑time captioning, potentially lowering manual editing costs.

Moreover, the inclusion of speaker diarisation and timestamps at the base price simplifies integration with analytics pipelines, enabling faster insight extraction from call recordings and video conferences.

Early adopters have reported that the combined benefits of higher accuracy and multilingual detection reduce the time spent on post‑processing by up to 30 %, allowing teams to focus on higher‑value tasks such as sentiment analysis and action‑item extraction.

Overall, SpaceXAI’s Grok Voice Transcribe 2.0 appears positioned to become a cornerstone technology for businesses seeking reliable, cost‑effective speech‑to‑text solutions across diverse linguistic environments.