OpenAI ships biggest ChatGPT overhaul: new app, 'Work' agent, browser, and GPT-5.6 family

OpenAI declared 'ChatGPT as you know it' over with a sweeping overhaul: a new desktop super-app, an agentic assistant branded ChatGPT Work that completes complex multi-step tasks across apps and files, a built-in browser, and the GPT-5.6 model family. The three tiers are named Sol (frontier reasoning for coding and research), Terra (high-capability everyday production work at lower cost), and Luna (fast, affordable high-volume tasks). The rollout spans the API, ChatGPT Plus, Team and Business.

On capability, OpenAI touted GPT-5.6 Sol proving a 50-year-old mathematical conjecture in under an hour by orchestrating 64 sub-agents — a marquee demonstration of the family's agentic reasoning. Alongside the models, OpenAI launched GPT-Live, a full-duplex voice architecture that listens and speaks simultaneously rather than taking turns, and returned ChatGPT to WhatsApp in the EEA.

The reception has been mixed. Axios and Inc. flagged a 'paradox of choice': the jumble of model names, reasoning levels and subscription tiers leaves users struggling to pick the right combination. Simon Willison publicly panned the 'AI employees' framing as disrespectful to humans and a misunderstanding of the tools. GPT-Live's conversational backchannels ('mhmm', 'yeah') drew complaints that it talks over users.

Competitively, the launch lands directly against Anthropic's Claude Sonnet 5 and xAI's Grok 4.5, both released the same week, and reframes the contest around agentic workflows rather than raw chat. Watch whether the agentic Work mode proves reliable enough for enterprise automation, and whether OpenAI simplifies the confusing tier lineup after the initial backlash.