OpenAI has rolled out two significant additions to its lineup: a new GPT-5.6 model family and a set of GPT-Live voice models built to power ChatGPT.
The headline feature of GPT-Live is that it can speak and listen at the same time. According to TechCrunch, OpenAI says this "full-duplex" ability is key for live translation and makes back-and-forth conversation feel more natural, rather than forcing users to wait for the AI to finish before responding. Several outlets, including MLQ.ai and HotHardware, describe the flagship as GPT-Live-1, and AI Insider frames voice as OpenAI's bet on the future interface for AI agents.
On the text side, OpenAI unveiled the GPT-5.6 family, which Techzine reports includes models named Sol, Terra, and Luna. CEO Sam Altman told CNBC the newest model is 54% more token efficient on agentic coding tasks. According to The Decoder, GPT-5.6 Sol scored 59 on the Artificial Analysis Intelligence Index—one point behind Anthropic's Claude Fable 5—while costing about $1.04 per task, roughly a third of what Anthropic's top model charges. The Decoder adds that Sol beats every competitor at agentic coding. Yellow.com reports a top OpenAI researcher claiming GPT-5.6 already beats human interns.
The launch also carried a political note: The Washington Post reports OpenAI released a system the Trump administration had "initially put on a leash," with qz.com and Yahoo saying GPT-5.6 Sol shipped after U.S. government approval. OpenAI also debuted a ChatGPT Work agent, per Constellation Research and MacRumors.
Why it matters: the releases push AI toward more human-like voice interaction while intensifying a price-and-performance battle with Anthropic that could reshape what AI tools cost.