On June 9, 2026, Google launched Gemini 3.5 Live Translate, an audio model that translates spoken conversation across more than 70 languages in near real time. It ships inside Google Translate, Google Meet, and a developer API. For businesses, real-time translation just moved from an expensive specialty service to a programmable feature you can build into everyday workflows.
That sounds like a consumer convenience, and for travelers it is. But the more consequential story is what happens when fluid, low-latency translation becomes cheap enough to embed in meetings, support queues, and sales calls. Language has always been a hard boundary on which markets a company can serve and which talent it can hire. That boundary is starting to move.
What Google Actually Shipped
According to Google's announcement, Gemini 3.5 Live Translate performs continuous speech-to-speech translation across more than 70 languages. The model auto-detects the language being spoken and generates translated speech that stays just a few seconds behind the speaker while preserving intonation, pacing, and pitch, so the output sounds like a person rather than a robotic readout.
It rolled out across several surfaces at once, according to reporting on the launch: the consumer Google Translate apps on Android and iOS worldwide with no sign-up, the Gemini Live API and Google AI Studio in public preview for developers, and Google Meet in private preview for select business Workspace customers. In Meet, support for more than 70 languages unlocks over 2,000 language combinations in a single meeting, a large jump from the small set of languages the product previously handled.
The pricing matters as much as the capability. Developer access through the Live API is inexpensive by historical standards, reported at roughly two cents per minute of audio, with token-based billing that varies by usage. For context, professional human interpretation is typically billed by the hour and booked in advance. When a capability gets that cheap, the spreadsheet math behind "should we offer this in another language?" changes completely.
Why This Is a Business Decision, Not a Gadget
It is tempting to file live translation under consumer features. That would be a mistake. The release follows OpenAI's own translation-capable voice models from May, part of a wider shift of voice AI crossing into production. Two of the largest AI vendors shipping near real-time, multilingual speech in the same quarter is not a coincidence. It signals that streaming speech translation has crossed from research demo into a commodity primitive.
When a capability becomes good and cheap at the same time, the strategic question flips. It stops being "can we technically do this?" and becomes "where in the business does this create value, and who moves first?" Most companies have quietly accepted language as a fixed cost: you serve the markets your staff can speak to, you hire from the talent pools you can interview, and you pay for interpreters only when the stakes justify it. Real-time translation at two cents a minute pressures every one of those assumptions.
Our take: the winners here will not be the companies that adopt the flashiest demo. They will be the ones that quietly remove language as a constraint from their existing operations, one workflow at a time.
Where Real-Time Translation Creates Value
The clearest near-term gains are in places where conversation, not documents, is the bottleneck.
Distributed and global teams. A meeting where everyone speaks their own language and understands everyone else is no longer a science-fiction scenario. For companies with engineering in one region, sales in another, and customers in a third, this lowers the friction of simply working together.
Multilingual customer support. Supporting customers in ten languages has traditionally meant hiring or outsourcing in ten languages. Real-time translation lets a smaller team cover far more ground. Paired with the context-aware AI customer service systems that already handle routine queries, translation extends reach without multiplying headcount.
International sales and field work. Sales and service conversations that previously required a local hire or a scheduled interpreter can now happen on demand. That shortens the path into new markets, where speed of first contact often decides who wins the deal.
Capturing that value is less about the model and more about plumbing: piping translated transcripts into your CRM, support desk, and meeting records so the output becomes searchable, auditable, and actionable. That integration work, not the translation itself, is where connecting AI into existing business workflows determines whether the capability sticks or stays a novelty. A translated conversation that disappears the moment the call ends is a missed opportunity; one that flows into your systems compounds over time.
What Real-Time Translation Does Not Solve
A realistic view requires naming the limits. Machine translation has improved dramatically, but it is not uniformly reliable, and the gap is widest exactly where the stakes are highest.
Accuracy varies by language pair. High-resource languages tend to perform far better than low-resource ones, so a capability advertised as "70+ languages" will not feel equally polished across all of them. Domain-specific terminology, legal and medical language, and culturally loaded nuance remain failure points. For anything contractual, regulated, or reputationally sensitive, a human in the loop is not optional.
There are also data considerations. Streaming live conversations through a third-party model means voice and content leave your environment, which is a question worth answering before you route sensitive calls through any provider. We are not offering legal or compliance guidance here; we are flagging that the decision deserves deliberate review rather than a default opt-in.
The honest framing is a tiered model: lean on AI for everyday scale, speed, and breadth, and reserve human experts for the conversations where a translation error carries real cost.
How Businesses Should Respond
You do not need a translation strategy. You need to find the one or two places where language is quietly costing you reach, then test whether this changes the equation.
- Find the language tax. Where are you losing deals, slowing support, or narrowing your hiring because of language? That is where to look first.
- Pilot narrowly. Pick a single workflow, such as a recurring multilingual standup or a support queue for one underserved language, and measure outcomes against your current approach.
- Build the integration, not just the demo. The value is in the transcript flowing into your systems. Treat translation as a feature inside your stack, not a separate app people have to remember to open.
- Decide the human-review line up front. Define which conversations can run fully automated and which always require expert review.
This is the same discipline that separates AI projects that ship from the ones that stall. As we have written about why most AI projects stall between pilot and production, the technology is rarely the hard part. Integration, governance, and clear ownership are.
Key Takeaways
- Google's Gemini 3.5 Live Translate, launched June 9, 2026, brings near real-time speech-to-speech translation in more than 70 languages to Translate, Meet, and a developer API.
- At roughly two cents per minute via the Live API, live translation has shifted from a specialty cost to a programmable feature, with OpenAI shipping comparable voice models the same quarter.
- The biggest gains come from removing language as a constraint inside existing workflows: global meetings, multilingual support, and international sales.
- Accuracy still varies by language and breaks down in high-stakes domains, so a tiered model with human review for critical communication remains the responsible default.
The businesses that move early on real-time AI translation will have a meaningful advantage in the markets they can reach and the talent they can hire. If you want to be one of them, let's start with a conversation.