GPT-5.6 Sol, Terra, and Luna — OpenAI’s Tri-Tier Model Means You Don’t Pay for More AI Than You Need

What Is GPT-5.6 and Why the Three-Tier Structure Matters

OpenAI released GPT-5.6 on July 9, 2026, but the real story isn’t the model — it’s how they’re selling it.

For the first time, OpenAI offers three permanent tiers under one model family: Sol, Terra, and Luna. Each targets a different level of reasoning complexity and cost.

Simon Willison described the launch as “a fundamental shift in how AI models are priced and consumed” in his July 9 analysis. Instead of one model that tries to do everything, OpenAI gave businesses a sliding scale: pay Sol rates for tasks that genuinely need frontier reasoning, and pay Luna rates for everything else.

For small business owners running multiple AI workflows, this changes the math on monthly AI spend. A single Luna query costs $0.001 per 1,000 tokens of output. A Sol query costs $0.03 per 1,000 tokens of output. If 80% of your tasks don’t need Sol-level reasoning, you’ve been overpaying.

Everyday Workloads Belong on Luna

Luna is the cheapest and fastest tier at $1 input / $6 output per million tokens. It handles high-volume, speed-sensitive tasks where a slightly less capable model still produces excellent results.

What Luna does well: email drafting, social media content, simple data extraction, customer FAQ responses, meeting summaries, classification tasks, and basic copywriting. OpenAI positions Luna for “speed and efficiency” — the tier that answers quickly enough to feel real-time in a customer-facing chatbot without costing $0.03 per response.

For a small business running 10,000 customer inquiries per month, Luna could cost roughly $100-200 in API fees depending on response length. The same volume on Sol would cost $500-1,000. That gap matters for businesses watching their tool stack costs.

Complex Work That Needs Terra

Terra sits in the middle at $2.50 input / $15 output per million tokens, offering “GPT-5.5-competitive performance at roughly half the cost of Sol,” per OpenAI’s pricing announcement. This is the everyday tier for most professional AI work.

Where Terra fits: detailed report writing, multi-step analysis, contract review, competitor research, grant writing, and marketing strategy documents. It gives you most of Sol’s reasoning depth at half the token cost.

For a business owner using AI to draft investor updates, analyze quarterly data, or write detailed SOPs, Terra is likely the sweet spot. You get the reasoning quality without paying premium Sol rates for every document.

When Only Sol Will Do

Sol is OpenAI’s current flagship at $5 input / $30 output per million tokens. This is the tier for complex reasoning, multi-step agent workflows, and tasks that genuinely need frontier capability.

Sol is overkill for drafting a blog post. It’s the right tool for: building custom AI agents that operate across multiple tools, analyzing a 200-page legal contract with embedded edge cases, debugging complex code across an entire codebase, or running a sophisticated data analysis with multiple conditional branches.

According to Vellum’s benchmarks published alongside the launch, Sol matches or exceeds GPT-5.5 on all major reasoning benchmarks while introducing better instruction following for complex agentic chains. But the cost is real — a single Sol agent call that processes 10,000 tokens output costs $0.30. The same call on Luna costs $0.06. The difference compounds fast at scale.

The 80/20 Rule of AI Spend for Small Business

The most practical takeaway for US small business owners: most AI use cases don’t require the most expensive tier.

A 2026 report from the Small Business & Entrepreneurship Council found that 82% of US small businesses now use AI tools, but 44% reported that AI costs went up unexpectedly in the past year. Much of that cost growth comes from paying premium per-token prices for simple tasks — classifying support tickets, generating product descriptions, answering basic customer questions.

GPT-5.6’s three-tier structure fixes this at the model level. A business that routes all email and social content through Luna, puts document drafting and analysis on Terra, and reserves Sol for the handful of weekly high-stakes tasks could cut AI API costs by 50-70% compared to running everything on Sol.

How the Tiers Compare

Real-World Use Cases: How US Small Businesses Should Tier Their AI

The theory is clear: use Luna for simple tasks, Terra for mid-complexity work, and Sol for the heavy stuff. But what does that look like for an actual US small business running daily AI workflows?

Consider a 12-person real estate agency using AI for three core tasks: drafting property listings, responding to buyer inquiries via chatbot, and analyzing quarterly market comparison reports.

The property listings are straightforward — 200-300 words of descriptive copy. This is Luna work. At Luna rates, 100 property listings per month cost roughly $2-4 in API fees. On Sol, those same listings would cost $15-25. The listings aren’t better on Sol. The agency saves 80% on a high-volume task and nobody notices the difference in quality.

The buyer chatbot handles 3,000 inquiries per month answering questions about square footage, school districts, HOA fees, and showings. This is also Luna territory — simple classification and lookup responses. Estimated cost on Luna: $15-30 per month. On Sol: $90-180. The chatbot answers faster on Luna because the model is lighter, so the customer experience actually improves.

The quarterly market analysis — comparing 50+ comps across three zip codes, identifying pricing trends, and generating a strategy memo — that is Terra work. Maybe one or two of those per month. Cost on Terra: $5-10. On Sol: $10-20. The difference is small at this volume, but over a year it adds up.

This business could run its entire AI stack for $25-45 per month in API fees using intelligent tier routing. Running everything on Sol would cost $120-230. The 60-80% savings doesn’t require reducing AI usage — it requires routing the right work to the right tier.

Another example: a US-based e-commerce brand with 500 SKUs uses AI to write product descriptions (500 per month), power a returns FAQ chatbot (2,000 inquiries per month), and generate weekly inventory analysis. Product descriptions on Luna cost $10-15 per month versus $50-80 on Sol. The chatbot on Luna costs $10-20 versus $60-120 on Sol. The weekly inventory analysis on Terra costs $5-8 per month versus $10-16 on Sol. Total savings: roughly 65%.

OpenAI’s own ChatGPT Work product already handles this routing automatically. But for businesses building custom AI workflows, the tier selection happens at the API call level. Every developer integrating GPT-5.6 should default to Luna and escalate to Terra or Sol only when the task requires it.

Prompt Caching Sweetens the Deal

GPT-5.6 also introduces updated prompt caching terms. Cache writes cost 1.25x the uncached input rate, but cache reads get a 90% discount. The minimum cache life is 30 minutes, meaning repeated queries with the same system prompt or context window — common in agent workflows — get substantially cheaper after the first call.

For a business running an AI receptionist that handles hundreds of daily calls with the same brand instructions and menu details, prompt caching turns the second through hundredth query into near-zero-cost operations on the cached prefix. This is where Luna with caching becomes genuinely cheap at scale.

What This Means for Your AI Stack in 2026

Small businesses using AI tools built on GPT-5.6 may or may not see these savings directly — it depends on whether the tool vendor passes the tier savings through to customers. But the model structure matters for anyone building custom AI workflows or evaluating API costs.

OpenAI’s application-level product, ChatGPT Work, launched alongside GPT-5.6 on July 9, uses the same model tiering. Tasks inside ChatGPT Work automatically route to the appropriate tier. A request to “summarize this email” hits Luna. A request to “analyze this 50-page lease agreement” routes to Sol.

The three-tier model is also a signal for where the industry is heading. Multiple model providers are moving toward tiered pricing: DeepSeek’s recent 75% price cut, Google’s Gemini tier structure, and Anthropic’s Claude pricing all reflect the same trend. The era of paying flagship prices for every AI query is ending.

For US small business owners, the practical advice is straightforward: audit what your AI tools are being used for. If every API call in your stack routes through a frontier-tier model, you’re wasting money. The tools are smart enough to tier themselves now — make sure your vendor is actually doing it.

Frequently Asked Questions