Grok Now Transcribes Calls That Mix English, Malay and Mandarin

xAI released Grok Voice Transcribe 2.0 on 18 September 2026, a speech to text model that is twice as accurate as its predecessor at the same price and ranks first for accuracy among 32 streaming transcription models. For a business in Malaysia, the standout feature is that it follows a conversation as it switches between English, Malay and Mandarin mid sentence.

What did xAI announce with Grok Voice Transcribe 2.0?

xAI built the new model on the same audio foundation model behind Grok Voice, which the company says already powers tens of thousands of customer support calls a day and transcribes millions of hours of video narration. On the public Artificial Analysis leaderboard, Grok Voice Transcribe 2.0 ranks first for accuracy among 32 streaming models. Internally, xAI tested it against telephony audio, conversations with Grok, spoken credentials such as account codes and email addresses, and short multilingual voice commands, and says it improved on version 1.0 across all four sets. Existing customers get the upgrade automatically, with no code changes needed, and pricing stays the same at 0.10 US dollars per hour of audio for batch transcription and 0.20 US dollars per hour for streaming.

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.from Sanchan Saxena, SVP of Teamwork Collection, Atlassian (xAI, 18 September 2026)

What can Grok Voice Transcribe 2.0 actually do?

• Transcribe a recorded file or an audio stream in real time, in batch or live.
• Label each speaker in the transcript automatically, at no extra cost.
• Transcribe up to 8 audio channels independently in one file.
• Give each word a start and end time with a confidence score.
• Accept up to 100 domain terms per request, such as product names, so unusual words are recognised correctly.
• Write out numbers, dates, currencies, phone numbers and email addresses in proper written form, and remove filler words such as um and uh.

Who could use this?

Business conversations in Malaysia rarely stay in one language, a customer might open in Malay, switch to English for a technical detail, and close in Mandarin. Most transcription tools built for a single language lose accuracy the moment that switch happens. xAI's own short phrase test, the closest to how someone actually speaks on a call, showed word error rate falling from 20.6 percent to 6.8 percent on this kind of multilingual speech. That is the difference between a customer service log a manager can trust and one that needs a person to redo it by hand. A retailer logging WhatsApp voice notes, a clinic recording patient calls, or a training provider captioning a workshop video in more than one language could all use this the same week it launched.

Also today

• ChatGPT for Word launched on 17 September 2026, bringing OpenAI's assistant into a sidebar inside Microsoft Word to draft, summarise, revise and format a document, on all ChatGPT plans including Free (OpenAI).
• Claude for Small Business added 43 workflows and 27 new integrations on 15 September 2026, including Shopify, Xero, Zoom and Gusto, after passing 900,000 installs since its May launch (Claude, Anthropic).

What to do this week

If your team already records customer calls, meetings or training sessions in more than one language, it is worth testing whether your current transcription tool actually keeps up when the language changes mid sentence, because most were never built for that. Our Master Claude AI for Business Productivity workshop covers exactly this kind of practical setup, using the right AI tool for daily tasks like meeting notes and customer records instead of forcing everything through one app.