OpenAI Launches GPT-5.6, Unifies ChatGPT and Codex, Introduces ChatGPT Work

OpenAI released GPT-5.6 with three tiers: Sol (flagship, $5/$30 per M tokens), Terra (mid-range, $2.50/$15), and Luna (fast, $1/$6). ChatGPT and Codex merged into a single desktop app with Chat, Work, and Codex modes. ChatGPT Work is an agent that connects to business apps (Slack, Gmail, Drive, etc.), autonomously executes multi-step tasks, and delivers finished documents, spreadsheets, or websites. Scheduled tasks, browser control with login support, and Sites (shareable web apps from conversations) were introduced. Pricing: Free (limited), Go ($8/mo), Plus ($20/mo), Pro ($100/$200/mo), Business, Enterprise. Work and Codex share an agentic usage quota separate from Chat; token-based consumption started in April. Codex has 5M weekly users, over 1M using it for non-dev tasks. OpenAI internal use: sales turned a customer call into a custom POC in 24h; finance cut month-end close from days to hours. A Claude Code migration tool was added. GPT-5.4 retires July 23. Ultra effort mode (Sol) uses parallel sub-agents, limited to Pro/Enterprise.
Key insights
GPT-5.6 Tiered Model Family Offers Performance and Price Flexibility
Sol targets complex reasoning and long autonomous work, Terra provides near-GPT-5.5 performance at half the price, and Luna prioritizes speed and low cost. This tiering allows users to choose the right balance of capability and expense for different tasks.
ChatGPT Work Transforms AI from Conversationalist to Autonomous Executor
Work connects to enterprise tools, breaks down complex projects, executes steps independently, and asks for input only when needed. It can run scheduled tasks, use a built-in browser, and deliver finished business artifacts, moving beyond chat-based assistance to full task completion.
Unified ChatGPT-Codex App Simplifies Product Line and Targets Broader Audience
Merging ChatGPT and Codex into one app with three modes eliminates product fragmentation. Codex's 5M weekly users, including over 1M non-developers, justified the unification. The app now serves both developers (code diffs, PRs) and business users (documents, slides) from a single interface.
Separate Quota Pools and Token-Based Pricing Align Cost with Usage Intensity
Chat messages have their own limits, while Work and Codex draw from a shared agentic quota based on token consumption. Heavier model choices and fast mode consume more quota. This structure ensures that intensive agent tasks are metered separately from simple chats, and users can buy extra credits.
OpenAI's Internal Adoption Demonstrates Significant Productivity Gains
Sales teams used Work to turn a discovery call into a custom POC in 24 hours (previously weeks). Finance teams compressed month-end close and forecasting from days to hours. These internal cases validate the agent's enterprise value and serve as proof points for customers.
Platform Play Positions OpenAI Against Office Suites Ahead of IPO
By combining chat, coding, browsing, file operations, and agentic automation into one app, OpenAI is building an AI-native productivity platform. With 40% of revenue from enterprises and an SEC filing for IPO, the unified super-app strategy aims to lock in business customers and compete with Google Workspace and Microsoft 365.
Codex Migration Tool Targets Anthropic's Developer Base
The addition of a Claude Code migration import feature, including settings and preferences, is a direct competitive move to attract developers from Anthropic's ecosystem, signaling intensifying rivalry in the AI coding assistant market.
Scheduled Tasks and Browser Control Extend Agent Autonomy
Work can run on schedules (daily briefings, monitoring), use a browser with login-state support and multi-tab operation, and even control desktop apps via Computer Use. This enables persistent, background automation that goes beyond single-session interactions.