Grok Voice Transcribe 2.0 Pricing Impact (2026)
xAI launches Grok Voice Transcribe 2.0 at the same REST and streaming rates as 1.0. See defaults, features, limits, and migration advice.
By AI Pricing Guru Editorial Team
AI Pricing Guru articles are maintained by the editorial workflow behind the site: daily pricing snapshots, provider source checks, and review passes for model launches, subscription limits, and billing changes.
TL;DR
- Grok Voice Transcribe 2.0 became available through xAI's Speech to Text API on September 17, 2026.
- xAI publishes the same REST and streaming rates for Transcribe 1.0 and 2.0; the live table below reads both from our pricing dataset.
- This is an opt-in upgrade: Transcribe 1.0 remains the default when the model parameter is omitted.
- Canary 2.0 on real accents, noise, names, numbers, diarization, and turn detection before moving production traffic.
Grok Voice Transcribe 1.0 vs 2.0 pricing
| Model | Primary / min | Primary / hour | Streaming / hour | Change |
|---|---|---|---|---|
| Grok Voice Transcribe 1.0 | $0.0017 | $0.1 | $0.2 | Baseline |
| Grok Voice Transcribe 2.0 | $0.0017 | $0.1 | $0.2 | No change |
Primary rates use the provider's published minute and hour values. Speech-to-text rows show REST as primary and list the separate WebSocket streaming rate where available. Additional text, tools, or telephony charges may apply.
Live values come from the canonical pricing API.
xAI released Grok Voice Transcribe 2.0 on September 17, 2026 for file and real-time speech-to-text workloads. The versioned model ID is grok-voice-transcribe-2.0, but xAI still routes requests without a model parameter to grok-voice-transcribe-1.0.
The pricing result is unusually simple: xAI’s rate card does not add a 2.0 premium. The build-time table above pulls the current REST and streaming rates for both versions from our maintained dataset, so it will reflect later changes without freezing launch-day prices in this article.
What changed in Grok Voice Transcribe 2.0
xAI now offers two pinned transcription models through the same Speech to Text API:
| Decision | Transcribe 1.0 | Transcribe 2.0 |
|---|---|---|
| xAI positioning | Original model | Best transcription model |
Default when model is omitted | Yes | No |
| File transcription | /v1/stt | /v1/stt |
| Real-time transcription | WebSocket STT | WebSocket STT |
| Published list-price difference | Baseline | None |
xAI has not published a version-to-version accuracy, latency, or language benchmark in the release note. “Best” is the provider’s description, not proof of a fixed improvement on every accent, codec, microphone, or call-center recording.
Pricing impact
Transcribe 2.0 arrives at the same published list rates as 1.0 for both REST and streaming. That means a controlled upgrade does not require a higher list-price budget at equal audio duration and mode.
Mode still matters. REST file transcription and live WebSocket streaming have separate rates, so moving a batch archive into a real-time pipeline can change the bill even when the model version does not. File downloads, storage, retries, observability, and downstream language-model work can also add costs outside xAI’s STT line item.
Use the live xAI API pricing page for the current catalog. If a transcript feeds summarization or extraction, price those text tokens separately with the AI token calculator. Teams comparing wider voice stacks should also check OpenAI API pricing and our earlier Grok Voice Think Fast 2.0 analysis; speech-to-text and speech-to-speech are different products.
What the API supports
The shared Speech to Text interface accepts audio files or public audio URLs and supports REST plus real-time WebSocket streaming. xAI documents 12 audio formats, files up to 500 MB, word-level timestamps, speaker diarization, key-term biasing, filler-word control, inverse text normalization, and up to eight channels.
For streaming, developers can request interim results, tune endpointing and voice-activity detection, send Opus packets, and enable Smart Turn so a pause is less likely to be mistaken for the end of a thought. Those controls can matter more than the model name in noisy telephony or dictation workflows.
Who benefits—and who should wait
Teams starting a new transcription integration can test 2.0 without paying a version premium. Existing users also get a low-risk canary path because 1.0 remains available and unchanged as the default.
Production systems that value reproducibility should not rely on the omitted-model default. Pin 1.0 while establishing a baseline, then send a measured traffic slice to 2.0. Wait for your own evidence if the workload includes regulated transcripts, uncommon names, mixed languages, heavy background noise, narrowband calls, or strict timestamp accuracy.
What developers should do now
- Set
modelexplicitly in both REST forms and WebSocket query strings. - Replay the same consented audio through 1.0 and 2.0; score word error rate, names, numbers, timestamps, diarization, formatting, and empty-audio behavior.
- Separate REST and streaming spend in dashboards because their billing modes differ.
- Keep language, key terms, VAD threshold, endpointing, Smart Turn, codec, and sample rate fixed during the comparison.
- For multipart REST requests, place options before
file; xAI warns that fields sent after the file may be ignored. - Promote 2.0 only when cost per accepted transcript improves after retries and human correction.
For an independent managed baseline, benchmark ElevenLabs on the same transcription set.
Affiliate disclosure: AI Pricing Guru may earn a commission from the sponsored link above at no extra cost to you. It does not affect this analysis.
Labs coverage
Grok Voice Transcribe 2.0 does not belong in the current AI Pricing Guru Labs leaderboard. That suite evaluates deterministic text answers, not audio transcription, timestamps, speakers, streaming endpointing, or correction effort.
A valid STT study needs licensed audio, pinned versions, disclosed codecs and sample rates, multiple languages and noise levels, word and entity error rates, latency, retries, human correction time, and billed cost per accepted transcript. Until that harness exists, we will not convert xAI’s “best” label into an unsupported score.
Bottom line
Grok Voice Transcribe 2.0 is an opt-in quality candidate with no published version premium. Pin the model, keep 1.0 as the control, and migrate only after 2.0 improves accepted transcripts on production-shaped audio.
Sources: xAI’s official release notes, Speech to Text documentation, and Voice API pricing. Pricing and documentation verified September 18, 2026.