OpenAI · updated 2026-08-29
Run GPT-5.6 (Sol / Terra / Luna): providers, pricing and flat-rate access
OpenAI's July 2026 generation, sold as three durable capability tiers: Sol (flagship reasoning), Terra (mid), and Luna (fast/cheap) — all with a 1.05M-token context window.
GPT-5.6 runs three ways: ChatGPT plans ($20–200/mo, app-and-Codex focused with usage limits), the OpenAI API per-token (Sol $5/$30 per M tokens, Terra $2/$12, Luna $0.20/$1.20 — long-context requests bill roughly double), or a flat-rate plan like Standard Compute, where all three tiers sit in the smart-routing pool at one fixed monthly price.
Your options, honestly
ChatGPT Plus / Pro$20–200/moFirst-party plans built around the app and Codex — usage-limited, not designed for arbitrary API agent traffic.
OpenAI APIPay-per-tokenSol $5/$30, Terra $2/$12, Luna $0.20/$1.20 per M tokens; requests past the short-context threshold bill about 2x. Agentic loops on Sol get expensive fast.
Standard Compute$39+/mo flatSol, Terra and Luna all in the routed pool — Luna handles routine steps, Sol-class the hard ones. Routing decides the tier; you can't pin one.
Azure / OpenRouterPay-per-tokenSame per-token economics under different billing; useful for enterprise procurement or provider fallback.
the flat-rate trade-off
Honestly: Luna alone is cheap enough per-token that a Luna-only workload should just pay per token. The flat-rate case is the mix — agents that want Luna speed on routine steps and Sol quality on hard ones without watching a meter, especially with long contexts where per-token rates double.
Running it in your agent
Any OpenAI-compatible agent (OpenCode, Cline, Aider, Roo, Continue) runs GPT-5.6 with a base-URL swap: api.stdcmpt.com/v1 for flat-rate, or OpenAI's endpoint per-token. On Standard Compute the 'standardcompute' model routes across all three tiers automatically.
FAQ
Sol, Terra or Luna for a coding agent?
Luna for routine edits and tool calls (it's 25x cheaper than Sol on input), Terra as the middle ground, Sol for hard multi-file reasoning. Most sessions are mostly-routine — which is why routed plans that make the split per-request tend to beat running Sol for everything.
Why did my GPT-5.6 bill jump mid-session?
Two compounding effects: agents re-send growing context every step, and long-context requests bill at roughly double rates (Sol goes from $5/$30 to $10/$45). Long agentic sessions hit both at once.
More: how flat-rate LLM APIs work · per-agent guides for OpenCode, Claude Code and Cline · model comparisons