On September 22, 2026, OpenAI launched GPT-6 as two models with opposite profiles: Luna and Sol. They’re not free and paid versions. They’re models with different architectures, optimized for different tasks, with pricing reflecting that difference.
This split changes how businesses should think about OpenAI. It’s no longer “which version of GPT do I use” but “which GPT model do I use for each step of my pipeline.”
Luna vs Sol: direct comparison
| Feature | GPT-6 Luna | GPT-6 Sol |
|---|---|---|
| Optimized for | Speed and volume | Reasoning and analysis |
| Input (per 1M tokens) | $0.10 | $2 |
| Output (per 1M tokens) | $0.50 | $10 |
| Cache reads | $0.01/M | $0.20/M |
| Context | 128K tokens | 272K tokens |
| Speed | Fast | Medium |
| ChatGPT Free | Yes (default model) | No |
Both cost approximately 50% less than their GPT-5.6 equivalents, at permanent list prices.
GPT-6 Luna: volume at minimum cost
Luna is the cheapest model on the market with competitive capabilities. At $0.10 per million input tokens, processing 10 million simple queries costs less than $1 in input.
Where Luna performs well
- Text classification: Categorize emails, tickets, messages by type, urgency or intent
- Simple data extraction: Pull specific fields from structured documents
- FAQ responses: Handle frequent queries with predefined answers
- Translation: General text without specialized terminology
- Short summaries: Condense texts of up to 2-3 pages
- Content generation: Drafts, variations, bulk commercial text
Where Luna falls short
- Analysis requiring multiple information sources
- Chain-of-thought reasoning
- Complex code with cross-file dependencies
- Tasks where errors are costly
GPT-6 Sol: reasoning at Sonnet pricing
Sol competes directly with Claude Sonnet 5.5 and Opus 5.5 in analytical capability. At the same price as Sonnet ($2/$10), it offers deep reasoning with OpenAI’s ecosystem.
Where Sol performs well
- Financial analysis: Interpret financial statements, detect anomalies
- Complex code: Multi-file refactoring, debugging with broad context
- Project planning: Decompose tasks, identify dependencies
- Technical writing: Precise documentation, detailed proposals
- Decision making: Evaluate options with multiple criteria
Luna + Sol strategy in enterprise
The most efficient combination uses both models in the same pipeline:
- Luna classifies each incoming ticket (category, urgency, language) → cost: pennies per thousands
- Luna responds to simple tickets with FAQ answers → minimal cost
- Sol analyzes complex tickets requiring investigation → cost only on cases that need it
- Luna generates the final response email based on Sol’s analysis → minimal cost
Result: 80% of tickets processed with Luna (~$0.60/M tokens), only 20% complex ones through Sol ($2/$10/M). Average cost per ticket drops 70% versus using Sol for everything.
Our recommendation
- Starting with OpenAI: GPT-6 Sol as primary model. Solid capability at $2/$10
- Already using GPT, want to reduce costs: Add Luna for 80% of simple queries. Direct savings
- Comparing with Claude: Sol competes with Sonnet 5.5 on price and capability. Difference is context (Sonnet: 1M, Sol: 272K) and native MCP (Claude). Details at Claude vs GPT-6 vs Gemini
- Need the cheapest on the market: Luna at $0.10/$0.50 has no rival on cost
For a complete overview, including Jev (cheaper still, but classification only) and Claude, see the AI model map October 2026.
Our artificial intelligence team works with all platforms. Direct integration with OpenAI API.
Request a free consultation and we’ll design the optimal architecture for your case.