OpenAI has overhauled ChatGPT's model lineup, replacing a year of naming confusion with three clearly defined tiers that map directly to job complexity and cost. The new system, which rolled out in July and August 2026, consolidates everything into the GPT-5 family with plain-language options: Instant for everyday chat, Thinking for harder problems, and Pro for research-grade work. Behind the scenes, three new flagship models named Sol, Terra, and Luna handle the heavy lifting, each optimized for a different balance of power, speed, and expense.

Why Did OpenAI Need to Rebuild Its Model Picker?

For most of 2025, using ChatGPT meant navigating a bewildering menu of overlapping options. Users could choose from GPT-4o for chat, GPT-4.1 for coding, GPT-4.5 for something in between, and then layer on o1, o3, o3-pro, and o4-mini for "reasoning" tasks that few people could fully explain. The confusion was so widespread that many users simply stuck with whatever loaded by default and never opened the model picker at all.

That fragmentation reflected OpenAI's rapid experimentation throughout 2025, but it created a real usability problem. The company spent the first half of 2026 consolidating, and by August 2026, the entire lineup had been reorganized around a single principle: match the model to the actual work, not to arbitrary version numbers.

What Are Sol, Terra, and Luna, and How Do They Compare?

The three new models represent a deliberate tiering strategy based on task complexity and cost efficiency. Sol is the flagship, built for long, multi-step agentic work that spans dozens of steps or hours of coding. It powers ChatGPT's reasoning experience for Plus and Pro subscribers after OpenAI merged the Instant and Thinking modes into one unified Sol-powered experience in August 2026. Terra sits in the middle, matching the quality of the previous flagship GPT-5.5 while costing roughly half as much to run through the API. Luna is the fast, cheap option, aimed at high-volume, simple work like summarizing, tagging, and basic drafting.

The pricing gap between them is substantial. Sol costs $5 per million input tokens and $30 per million output tokens. Terra runs at $2 per million input tokens and $12 per million output tokens. Luna, after a steep July 2026 price cut, costs just $0.20 per million input tokens and $1.20 per million output tokens. That means Luna is 25 times cheaper than Sol for input processing.

How to Choose the Right Model for Your Task

• Casual chat and quick questions: Instant covers everyday writing help and day-to-day requests without any setup. It is the default for free-tier users for a reason.
• Long documents and business writing: Thinking mode or Sol on paid plans delivers noticeably more careful, structured output for spreadsheets and work that needs a second pass.
• Software engineering and multi-step coding: Sol and its coding-focused predecessors earn their keep for developers building agentic tools, with consistently better follow-through on multi-step tasks compared with earlier GPT-5 versions.
• Bulk content work at scale: Luna gets categorization, summarization, and short product summaries done fast and at a fraction of the cost of the flagship.
• Voice and live interaction: OpenAI split this out entirely, moving real-time voice conversations to a dedicated live-interaction model rather than bundling it with text-based tiers.

For teams building on the API rather than just using the app, this tiering structure solves a real operational problem. The old lineup made it impractical to mix models within a single product, but the new pricing makes it economical to route the easy 80 percent of requests to Luna, send anything ambiguous to Terra, and reserve Sol for the handful of tasks that actually need frontier reasoning.

What New Features Come With Sol?

Sol ships with two features aimed squarely at teams doing complex work. Ultra mode runs several subagents on a hard task in parallel instead of working through it step by step alone. Programmatic Tool Calling lets the model write its own short scripts to chain tools together rather than waiting on manual instructions for each step. Both features are designed to reduce the back-and-forth that slows down long research sessions or multi-file coding projects.

How Much Does Each Tier Cost for Individual Users?

The subscription math for individual users differs from API pricing. Free access costs nothing and runs on Luna with basic limits. The Go tier runs about $8 per month with higher caps. Plus, at $20 per month, is where Sol shows up for individual users, alongside expanded research tools and image generation. All current models can read images and uploaded files, and most can generate images inline.

The shift from version numbers to planetary names reflects a company betting that most people do not want to memorize version numbers. They want a size that fits the job. By August 2026, that philosophy had become the organizing principle of ChatGPT's entire interface, making it easier for both casual users and teams to pick the right tool without getting lost in technical specifications.

" }