May 19, 2026
8×8 Taps OpenAI’s Latest Voice AI Model to Make Agents More Reliable
8×8 has added support for OpenAI’s GPT Realtime 2 inside 8×8 AI Studio, giving businesses running live voice agents a more dependable engine for real-time customer conversations. The update is available now to customers in 8×8 AI Studio’s early availability programme, while existing agents will continue running on their current model until teams choose to switch.
Hunter Middleton, Chief Product Officer at 8×8, said: “When a voice agent conversation doesn’t go as planned, it can’t pull up the account, drops the transfer, loses the thread, the customer feels it immediately. This update targets these failure modes that matter in live customer interactions.”
A Smarter Engine Under the Hood
At the core of the update is GPT Realtime 2 which brings GPT-5-class reasoning and a 128K context window, allowing AI agents to follow longer, more complex conversations without losing track. A customer who needs their account verified, inventory checked, and the case escalated to a specialist should get through that sequence with fewer dropped steps along the way.
8×8 has also introduced a per-agent reasoning effort control. Most customer interactions will work well at the default setting, but teams can turn the dial up for workflows that involve more steps and more tool usage. The tradeoff is added response time, which means teams can decide for themselves whether speed or depth is the right priority for a given use case.
Transcription Without Manual Adjustment
Every voice session in AI Studio now defaults to Realtime-Whisper, OpenAI’s latest transcription model. The result is cleaner records in the call log and in the live adviser interface, with no need for teams to change any settings. Accurate transcription has a cascading effect: supervisors can review conversations with confidence, AI agents have the full picture when picking up a conversation mid-flow, and businesses can act on what customers actually said rather than an approximation of it.
As more organisations deploy voice agents in production, the quality of the underlying transcript determines how much value the rest of the system can extract from every interaction.
No Disruption, Opt-in Only
8×8 has designed the rollout so that nothing changes for teams that are not ready to move. Agents already in production keep running on their existing configuration. When a team does decide to switch models through the agent editor, AI Studio automatically substitutes a compatible voice if the current one is not supported and logs the change, all without interrupting live workflows.
The update lands at a moment when executive pressure to deploy AI in customer service is intensifying. Research from Gartner found that 91 percent of customer service leaders reported pressure from leadership to implement AI in 2026, with improving satisfaction, operational efficiency, and self-service success identified as the top priorities. For teams already running voice agents, the question is no longer whether to use AI but whether the underlying model is reliable enough for the interactions that matter most. With GPT Realtime 2, 8×8 is making the case that the answer just got closer to yes.
