$5 / $30
Sol, the frontier tier: hardest reasoning, agentic work, the requests where failure is expensive
$2.50 / $15
Terra, the mid tier: production default for most chat, coding, and generation traffic
$1 / $6
Luna, the volume tier: classification, extraction, short summaries at scale
Sol, Terra, and Luna are the three capability tiers of OpenAI's GPT-5.6 family, launched July 9, 2026. Sol is the frontier model ($5 input / $30 output per million tokens), Terra the production mid-tier ($2.50/$15), and Luna the volume tier ($1/$6). All three share a 1.05M-token context window and 128K max output; they differ in reasoning capability, not features.
What actually differs between the tiers
Capability, not surface area. All three take the same API calls, the same context length, the same tool-calling and streaming. Sol reasons deepest and costs the most per token; Luna is fastest and cheapest and falls over first on hard problems; Terra sits between. If you were hoping one tier had a secret feature the others lack: no. The decision is purely quality-per-dollar for each slice of your traffic.
The price math, side by side
| Tier | Input $/M | Output $/M | A code-gen request* | A short extraction* | Anthropic's rung |
|---|---|---|---|---|---|
| GPT-5.6 Sol | $5.00 | $30.00 | $0.0815 | $0.0087 | Opus 5 ($5/$25, cheaper output) |
| GPT-5.6 Terra | $2.50 | $15.00 | $0.0408 | $0.0044 | Sonnet 5 ($2/$10 now standard; increase cancelled) |
| GPT-5.6 Luna | $1.00 | $6.00 | $0.0163 | $0.0017 | Haiku 4.5 ($1/$5, cheaper output) |
*List-price math on our measured task token profiles (100/2,700 code-gen, 180/260 extraction); we meter Anthropic models directly and publish those bills in the pricing guide. Caching: reads at 10% of input across the family; cache writes cost 1.25x. Batch API halves both sides. Prices: OpenAI pricing reference, verified 2026-07-28.
Which traffic belongs where
Luna: classification, routing decisions, extraction, short summaries, anything you run thousands of times daily with a clear success check. Terra: the default for everything else: chat products, routine code generation, drafting. If you only pick one tier, pick Terra and escalate failures. Sol: the requests where a wrong answer costs real money or engineering time: hard debugging, architectural reasoning, agentic sessions with many dependent steps. The expensive mistake is running Sol as the default: on our measured task profiles it costs 5x Luna on work Luna handles.
The trap in tier-splitting by hand
The honest catch: assigning traffic to tiers is a standing job, not a one-time decision. Failure rates move, prices change (Anthropic's mirror-image ladder undercuts each rung, and their mid-tier price changes September 1), and every failed cheap run bills you twice. That per-request assignment problem is exactly what a blended model automates; ours publishes published benchmark scores so the claim is checkable. If you would rather keep the assignment manual, the cost calculator prices any split across all three tiers and their Anthropic rivals.
The exact questions people search
What is the difference between GPT-5.6 Sol, Terra, and Luna?
Capability tiers of the same family: Sol is the frontier reasoning model at $5/$30 per million tokens, Terra the production mid-tier at $2.50/$15, Luna the volume tier at $1/$6. Same API, same 1.05M context, same features; they differ in how hard a problem they can solve reliably.
Is GPT-5.6 Sol worth it over Terra?
For hard reasoning, agentic chains, and high-cost-of-failure requests, usually. For routine chat and code generation, Terra at half the price handles most traffic; the honest pattern is Terra by default with escalation to Sol on failure.
Which is cheaper, Luna or Claude Haiku 4.5?
Haiku, slightly: $5 versus $6 per million output tokens with input tied at $1. On a short extraction task that is roughly $0.0016 versus $0.0017. At volume the gap is real money; quality on your specific traffic decides it.
Do Sol, Terra, and Luna have different context windows?
No: the whole GPT-5.6 family shares the 1.05M-token context window and 128K maximum output. Long context is a family feature, not a tier feature.
Compare any two models
VS
List rates and dated competitor measurements: prices and measured bills. Pareto 26.9 measured task costs are not published. Verbosity: Edition 2.
Or stop assigning tiers by hand. $100 to verify.
Get started with Unbiased
Don’t act on this yourself. Hand it to your agent and let it do the switching math for you.