GPT-5.6 Public: Sol/Terra/Luna Exit the Trusted-Partner Gate
OpenAI opened GPT-5.6 Sol, Terra, and Luna to public access after a June 26 government-coordinated trusted-partner preview. AA puts Sol max at 59 (1 below Fable 5) at ~1/3 Fable cost-per-task—intelligence-vs-cost Pareto, staged by Washington.
GPT‑5.6’s two-week government-gated preview—and July 9 public unlock—cements a new U.S. pattern after the Fable freeze: frontier launches can be staged by Washington. Opening Sol/Terra/Luna to everyone is both a competitive answer to Anthropic’s restored Fable wave and a test of whether “voluntary” pre-release review remains temporary.
OpenAI on July 9, 2026 publicly launched the GPT‑5.6 family—Sol, Terra, and Luna—ending the June 26 limited preview that restricted access to a small set of government-shared trusted partners. Company posts on July 8 confirmed public launch “this Thursday” with global preview expansion already underway; press reported Commerce/White House clearance for broader release after additional testing.
From gated preview to public
- June 26: OpenAI “Previewing GPT‑5.6 Sol” — limited API/Codex access for trusted partners at U.S. government request; company objected to permanent gating.
- July 8: OpenAI X / Altman posts — Sol/Terra/Luna public launch Thursday (July 9); expand preview access globally now.
- July 9: Public availability day for the three-model series (ChatGPT, Codex, and API breadth per contemporaneous rollout reporting).
Model tiers (from June 26 primary + Engadget/Neowin July coverage):
| Model | Role | Pricing notes (reported) |
|---|---|---|
| Sol | Flagship / most capable | ~$5 / $30 per MTok input/output (reported) |
| Terra | Balanced everyday | Competitive with GPT‑5.5 at ~half cost; ~$2.50 / $15 |
| Luna | Fast, lowest cost | ~$1 / $6 |
Exact plan placement in ChatGPT vs API may phase during rollout day; the structural story is three durable tiers plus exit from the emergency trusted-partner gate.
Claims vs checks
| Source | Snapshot | Standing |
|---|---|---|
| Artificial Analysis Intelligence Index | AA article Jul 9, 2026 | Sol (max) = 59 — 1 point below Claude Fable 5 (max); |
| AA Coding Agent Index | Same | Sol (max) in Codex = 80 — leads DeepSWE, Terminal-Bench v2, SWE-Atlas-QnA (ties Grok 4.5 in Grok Build on SWE-Atlas-QnA); cheaper per task than Fable/Opus in Claude Code harness comparisons |
| AA-Briefcase | Same | Sol second to Fable overall; highest Presentation Elo of any model; Fable still leads rubric / analytical quality Elo |
| Arena Text Overall | Board dated Jul 16, 2026 | gpt-5.6-sol-xhigh ~#10 at 1486 ±9 (~4,095 votes) — same score band as kimi-k3; Fable 5 still #1 at 1507 |
| Design Arena | Official results (launch week) | Sol #1 overall at Elo 1353 (above Fable 5 on that design board per Design Arena posts) |
Vendor Terminal-Bench claims (e.g. Sol 88.8%) sit beside these boards: AA and Arena are the independent cross-checks enterprises cite when comparing Sol to Fable and Grok.
Limits
- Token prices above are contemporaneous reportage; confirm against OpenAI’s live pricing table.
- AA and Arena snapshots move; Jul 9 / Jul 16 boards are dated windows.
- ChatGPT vs API plan placement may still phase on rollout day.
- Government clearance narrative is press/company framing of the trusted-partner exit—not a published Commerce order text in the sources above.
Sources
- OpenAI: “Previewing GPT‑5.6 Sol” (June 26, 2026) for model family and gated-preview framing.
- OpenAI / Sam Altman X posts confirming July 9 public launch (July 8, 2026).
- Artificial Analysis: “GPT-5.6 benchmarks across Intelligence, Speed and Cost” (July 9, 2026).
- Arena.ai Text Overall leaderboard snapshot (July 16, 2026); Design Arena official Sol #1 results (launch week).
- Engadget, Neowin, PYMNTS, Reuters video coverage of July 9 public rollout after government clearance (July 8–9, 2026).