Industry·
OpenAI Launches Live Dictation for ChatGPT with Unlimited Recording
ChatGPT now transcribes speech in real time, drops the 10-minute cap, and rolls out across all platforms — signaling growing demand for low-latency speech inference at scale.

OpenAI has rolled out Live Dictation for ChatGPT, letting users speak prompts and see text appear word-by-word as they talk. The update removes the previous 10-minute recording ceiling and is available on every tier — free, Plus, Team, Enterprise, and Edu — across web, desktop, and mobile apps.
Real-time transcription changes the interaction model
Instead of recording a clip, waiting for upload, then receiving a transcript, the new waveform view streams tokens back to the prompt box instantly. Users can edit the live text before committing it, turning voice into a first-class input modality rather than a convenience feature.
OpenAI also noted reliability improvements behind the scenes. The change follows this year's overhaul of ChatGPT's two-way voice mode and sits alongside the still-beta "record mode" for meeting transcription, which remains gated to paid workspaces.
Why it matters for GPU / AI infrastructure
Streaming speech-to-text at consumer scale demands continuous, low-latency inference on GPU clusters — especially when millions of concurrent sessions must stay under 200 ms end-to-end. Each live dictation session occupies a persistent model instance, driving sustained compute density rather than bursty batch workloads. For cloud providers and hardware buyers, this signals a shift toward always-on audio pipelines that favor high-throughput, memory-efficient GPUs and optimized serving stacks (TensorRT-LLM, vLLM, custom kernels). Capacity planning must now account for steady-state voice traffic alongside traditional chat and code workloads.
- aigpu
- ai gpu
- ai gpu cloud
- aigpu dubai
- openai
- speech-to-text
- gpu inference
- real-time ai
By AiGpu Editorial · Editorial rewrite based on public reporting (The Verge AI)
← All articles