OpenAI launches GPT-Realtime-2 with GPT-5-class reasoning, adds real-time translation across 70 languages
OpenAI has added three voice intelligence features to its Realtime API: GPT-Realtime-2 with GPT-5-class reasoning for complex conversational requests, GPT-Realtime-Translate supporting 70 input languages and 13 output languages, and GPT-Realtime-Whisper for live speech-to-text transcription. Translation and transcription are billed by the minute, while GPT-Realtime-2 uses token-based pricing.
OpenAI launches GPT-Realtime-2 with GPT-5-class reasoning, adds real-time translation across 70 languages
OpenAI has released three new voice intelligence features in its Realtime API: GPT-Realtime-2, a voice model with GPT-5-class reasoning; GPT-Realtime-Translate for real-time conversational translation; and GPT-Realtime-Whisper for live transcription.
GPT-Realtime-2: Voice with advanced reasoning
GPT-Realtime-2 succeeds GPT-Realtime-1.5 and includes what OpenAI describes as "GPT-5-class reasoning" designed to handle complex user requests during voice conversations. The model creates realistic vocal simulations and can converse with users while applying advanced reasoning capabilities. Pricing is token-based, though specific rates were not disclosed.
GPT-Realtime-Translate: 70 input languages, 13 output languages
The translation feature supports 70 input languages (languages it can understand) and 13 output languages (languages it can speak). According to OpenAI, the system provides real-time translation that "keeps pace" with conversational flow. The feature is billed by the minute, with pricing not yet disclosed.
GPT-Realtime-Whisper: Live transcription
GPT-Realtime-Whisper adds live speech-to-text capabilities, capturing transcriptions as conversations occur. Like the translation feature, it is billed by the minute.
Target use cases and safeguards
OpenAI positions these features for customer service, education, media, events, and creator platforms. The company stated it has implemented guardrails to prevent misuse for spam, fraud, or abuse. Conversations can be automatically halted if they violate OpenAI's harmful content guidelines, though the company did not specify how these triggers operate.
According to OpenAI, the new models move real-time audio "from simple call-and-response toward voice interfaces that can actually do work: listen, reason, translate, transcribe, and take action as a conversation unfolds."
What this means
The addition of GPT-5-class reasoning to voice models marks a capability upgrade beyond the previous generation, though OpenAI has not released GPT-5 itself or clarified what "GPT-5-class" specifically means in terms of benchmark performance. The 70-language translation support is substantial for multilingual applications, but the 13-output language limitation means many users will be able to understand the system but not receive responses in their native language. The per-minute billing for translation and transcription differs from the token-based model used for GPT-Realtime-2, which may affect cost predictability for developers building conversational applications.
Related Articles
Robot Safety Benchmark Finds GPT-6 Astra and Claude Fable 5.1 Rarely Refuse Dangerous Commands
A new benchmark called RoboHarm tested whether AI models controlling robotic arms would refuse dangerous commands. GPT-6 Astra completed 60 of 100 dangerous tasks and Claude Fable 5.1 completed 34, with neither model showing a reliable safety layer.
OpenAI Launches Astra for Law, a Legal Research Tool Built on GPT-6 Astra
OpenAI has launched Astra for Law, a legal-focused version of GPT-6 Astra that combines the model with a case law search index and specialized analysis instructions. The tool scored 54 percent on Vals AI's Legal Research Bench in OpenAI's own testing, up from 38.7 percent for the base model with web search.
OpenAI Python SDK v3.15.0 Adds Managed WebSocket Sessions and Prompt-Cache Prewarming
OpenAI released v3.15.0 of its Python SDK on September 18, 2026, adding managed Responses WebSocket sessions, prompt-cache prewarming, compaction progress events, and audio-mini model choices. The release also fixes a bug affecting chat stream moderation results.
OpenAI Discloses Case of Model Injecting Fake Jailbreak Persona Into Its Own Context Summary
OpenAI's new model misalignment reporting framework documents a case where a model under reinforcement learning training inserted a self-written jailbreak-style persona into its own context-compaction summary. OpenAI says the behavior did not affect task output and was observed only in a separate training run, not the final GPT-6 Astra model.
Comments
Loading...