GPT-5.6 is live for everyone — the gate is gone

Quick summary

• Three tiers: Sol (flagship), Terra (balanced, roughly GPT-5.5 quality at about half the cost), Luna (fast and cheap).
• Live since July 9, 2026 — ChatGPT, the API, and Codex, after US Commerce Department clearance ended the ~20-partner gate. Plus/Pro/Business/Enterprise plans get Sol at medium and higher effort.
• Confirmed API pricing: Sol $5/$30, Terra $2.50/$15, Luna $1/$6 per million input/output tokens — Sol is half Fable 5's $10/$50.
• OpenAI says this kind of government sign-off on releases shouldn't become “the long-term default.”

Who can use GPT-5.6 right now

Access is limited to roughly 20 partner organizations whose participation was vetted and shared with the US government, and they reach the models through the API and Codex — not through consumer ChatGPT. OpenAI has not disclosed the partner names.

If you're a regular ChatGPT user or a typical API developer, you cannot use GPT-5.6 yet. Be careful with lists naming specific companies — some names circulating actually belong to Anthropic's separate Mythos program, not OpenAI's.

When everyone else gets it

OpenAI says it plans to make Sol, Terra, and Luna generally available “in the coming weeks,” across ChatGPT, the API, and Codex, and that it will publish an updated system card at that point. There is no specific calendar date for consumer ChatGPT or for general API access; reporting says more customers are expected to be added within the following week.

• How Sol compares to Fable 5 and Mythos 5 The benchmark read, with OpenAI's own numbers
• Latest Fable 5 and access signals Confirmed, official, and community — kept separate

The three tiers: Sol, Terra, Luna

GPT-5.6 isn't one model — it's three fixed capability tiers, named after the sun, earth, and moon.

• Sol — the flagshipFor hard reasoning, long-horizon agentic work, and coding. It adds two new settings: a max mode that lets it reason longer on one chain, and an ultra mode that splits a task across sub-agents to run in parallel.
• Terra — the everyday tierDescribed as GPT-5.5-competitive at roughly half the cost. The one most teams running real volume will reach for.
• Luna — fast and cheapThe lowest-cost, highest-throughput option, for routine high-volume tasks.

Pricing

Here's what each tier costs. Terra is the headline value — close to last-generation flagship quality at roughly half the price.

Per million tokens (input / output), as reported across multiple outlets attributing OpenAI. A Cerebras option (up to ~750 tokens/second) is cited for July but is vendor-reported, not first-party-confirmed.

Performance — and how to read it

OpenAI's launch leads with Terminal-Bench 2.1, an agentic-coding benchmark. On OpenAI's own chart, Sol's multi-agent ultra mode leads, standard Sol edges Claude's Mythos 5, and Fable 5 (now back, as of July 1) sits a few points back:

Important: these four model scores are OpenAI-reported, not independently verified — public leaderboards don't yet list them, and their top independently-measured score sits well below Sol's claimed 91.9. Only Gemini 3.1 Pro's 70.7 is independently confirmed. There's no published GPT-5.6 ExploitBench number either; OpenAI only claims Sol matches Mythos on cybersecurity while using about a third of the tokens.

How it compares to Claude — Fable 5 and Mythos 5

The headline is that Sol beats Claude's frontier models on OpenAI's chart — but the gap is narrower than it looks. Standard Sol (88.8) and Mythos 5 (88.0) are within a point; Sol Ultra's 91.9 comes from a mode that spends far more compute. And the access asymmetry is over: Fable 5 is runnable, and GPT-5.6 went public July 9 — you can benchmark both on your own work today. At confirmed prices, Sol costs half of Fable 5 per token; whether it holds up on your tasks is exactly what the days before July 19 are for.

• Full comparison: Sol vs Fable 5 vs Mythos 5 The chart, the caveats, and the honest read
• What you can actually run today Model-by-model options, including the now-back Fable 5

Why the rollout is contested

The limit came at the request of the US government, tied to a June executive order, and is being handled customer by customer. OpenAI's own statement pushes back: it says this kind of government access process shouldn't become the long-term default, because it keeps the best tools from developers, enterprises, cyber-defenders, and global partners.

Sam Altman told staff the company has “made clear to the U.S. government that this is not our preferred long term model, and will work with them and others in industry to achieve a more sustainable approach for future releases.” Commentary has been blunter — OpenAI strategy lead Dean Ball said US AI policy has gone “from implausibly libertarian to increasingly draconian and opaque,” while on Hacker News the loudest reactions were “regulatory capture in action” and “so much for the open in OpenAI.”

Why this is on a Fable 5 site

If you came here because of Claude Fable 5, GPT-5.6 is part of the same story in two ways. First, it's the new performance reference point — Sol's headline benchmark is measured against Mythos 5 and Fable 5. Second, it shows how the access story split: Fable 5 was suspended and is now back (as of July 1) after the US lifted its export controls, while GPT-5.6 is still behind a US-government access wall. Same kind of government restriction, opposite outcomes this week.

From the Field Guide

Stop guessing which model to use

Weekly verified workflows for the credits era: which model in which seat, what it costs per task, and copy-ready prompts — plus an 8-runbook Starter Pack you can download once and keep. Every claim checked against published pricing, sources shown.