OpenAI Cut GPT-5.6 Luna 80%: What It Changes for Agents

On July 30, 2026, OpenAI published Advancing the price-performance frontier with GPT-5.6. The headline move is Luna: from $1.00 / $6.00 per 1M tokens down to $0.20 / $1.20 — an 80% cut. Terra dropped 20% to $2 / $12. Sol is unchanged.

That matters for agent operators more than for chatbot hobbyists. Agents burn tokens on tool loops, long contexts, and scheduled runs. A model that was “almost cheap enough” overnight became the default candidate for high-volume duties.

The numbers (official list)

Confirm on OpenAI’s pricing docs before you quote a budget. Official list as of July 30:

• GPT-5.6 Luna (gpt-5.6-luna) — $0.20 input / $1.20 output per 1M · cached input $0.02 · ~1.05M context
• GPT-5.6 Terra (gpt-5.6-terra) — $2.00 / $12.00 · cached $0.20
• GPT-5.6 Sol (gpt-5.6-sol) — unchanged (still the flagship / expensive lane)

Gotcha: gpt-5.6 is Sol, not Luna

The bare alias gpt-5.6 routes to Sol. Teams who “upgrade to 5.6” without pinning gpt-5.6-luna can quietly pay flagship rates. If your goal is the cheap tier, set the model id explicitly.

What changes for agent cost

At $0.20/$1.20, Luna sits in the same neighborhood as mini/flash/nano SKUs — with a much larger context window and the GPT-5.6 tool surface (including programmatic tool calling on the Responses API). For CloudAxis-style work — browser duties, scheduled monitors, multi-step specialists — that is the interesting part: you can afford more tool rounds per dollar.

Caveats still apply. Independent testing has reported Luna’s long-context recall degrading well before the full 1M window, and vendor benchmarks should be read with the usual skepticism. Cheap is not the same as “always pick Luna.”

We updated our live tools today

Stale prices on ranking tools are worse than no tools. Same-day updates:

• AI model pricing tracker — recollected catalog; GPT-5.6 family uses OpenAI official list when aggregators lag
• Cheapest LLM API guide — rewritten around the Luna cut
• AI cost calculator (CloudAxis) and CloudyBot calculator — Luna, Terra, and Sol added as first-class models

ChatGPT access vs API

On ChatGPT Work and Codex, Free and Go users get Terra; selecting Luna requires Plus or above (per OpenAI’s July 30 post). The API has no plan gate — any account can call gpt-5.6-luna and pay list rates. That split is why “Luna without Plus” searches will spike even though free ChatGPT users already see Terra in some surfaces.

CloudAxis Auto Mode is powered by GPT-5.6 Luna on every plan including Free — no manual picker. This post covers the price shock and the tools; product Auto Mode now runs on Luna.

Frequently asked questions

What is the new GPT-5.6 Luna API price?

$0.20 per 1M input tokens and $1.20 per 1M output tokens, with cached input at $0.02 — an 80% cut from the previous $1/$6 list, effective July 30, 2026.

Did Terra and Sol also get cheaper?

Terra dropped 20% to $2/$12 per 1M. Sol pricing was unchanged.

Is gpt-5.6 the same as Luna?

No. The gpt-5.6 alias routes to Sol (flagship). Use gpt-5.6-luna explicitly for the cheap tier.

Can I model Luna spend on CloudAxis?

Yes — the AI cost calculator on cloudaxis.ai now includes GPT-5.6 Luna, Terra, and Sol with the July 30 list prices. Pair it with CloudyBot’s live pricing table for catalog-wide comparisons.