Meet ChatGPT Work and GPT-5.6: OpenAI's Cosmic Tiers Unleash Long-Horizon AI Agents

The chatbot era is officially drawing to a close, replaced by the dawn of the autonomous workplace agent. In a massive global release, OpenAI has launched its next-generation GPT-5.6 model family, abandoning the old "Turbo" and "Mini" naming schemes in favor of a cosmic, tier-based naming structure: Sol, Terra, and Luna.

Launched alongside this new model generation is ChatGPT Work, an agentic desktop productivity workspace that integrates OpenAI’s advanced coding engine, Codex, and a built-in multi-tab browser directly into a unified "superapp". This powerhouse combination is designed to run complex, long-horizon tasks for hours, completely on autopilot.

The Celestial Tiers: Sol, Terra, and Luna

Rather than constantly rotating version numbers, OpenAI's 5.6 generation introduces three permanent capability tiers. The version number (5.6) defines the underlying model architecture, while the names represent durable capability baselines that can be updated independently over time.

• Sol (The Flagship): Engineered for maximum intelligence and complex reasoning. Sol is built specifically to handle difficult coding, security research, and long-horizon tasks that run autonomously for hours.
• Terra (The Balanced Middle): This is the everyday professional workhorse, positioned to deliver performance competitive with the previous GPT-5.5 generation but at half the price.
• Luna (The Fast & Affordable): Designed for high-volume, low-latency background workflows like bulk data sorting, RAG preprocessing, and rapid-fire triage.

+-------------------------------------------------------------------------+
|                              GPT-5.6 Family                             |
+-------------------+--------------------------------+--------------------+
| Tier              | Core Target                    | API Cost (1M Tok)  |
+-------------------+--------------------------------+--------------------+
| ☀️ Sol            | Hard Coding & Complex Agents   | $5.00 In / $30 Out |
| 🌍 Terra          | Daily Work & General Tasks     | $2.50 In / $15 Out |
| 🌙 Luna           | Speed & Low-Cost Automation    | $1.00 In / $6 Out  |
+-------------------+--------------------------------+--------------------+

From "Answering" to "Finishing" with ChatGPT Work

The core engine of this release is ChatGPT Work, OpenAI's answer to autonomous enterprise productivity. Instead of merely responding to a prompt, the agent accepts high-level objectives and delegates them to sub-agents. It coordinates with files, desktop environments, and web tools via a brand-new Ultra Mode—which runs up to four sub-agents in parallel—and Programmatic Tool Calling, allowing the AI to generate and execute its own JavaScript locally in an isolated runtime.

Whether you need a fully populated database, a multi-slide presentation, or an interactive prototype, ChatGPT Work executes the plan, allowing users to schedule and monitor recurring, multi-step actions directly from their desktop or mobile device.

Beating the Competition on Long-Horizon Workflows

OpenAI’s benchmarking highlights a major leap in agent durability. On Agents' Last Exam—a grueling benchmark designed to test long-running, multi-step professional tasks across 55 fields—GPT-5.6 Sol scored 53.6, outperforming Anthropic's Claude Fable 5 by 13.1 points. Additionally, Sol leading the Artificial Analysis Coding Agent Index at 80 points makes it an incredibly robust option for autonomous repository maintenance and software engineering.

For teams looking to scale, the smart play is already shifting toward multi-model routing. Developers are using Luna for cheap, constant background checks, Terra for daily chats, and reserving Sol for deep logical escalations. This paradigm shift is rapidly lowering token overhead while keeping agentic execution at an all-time high.