June 26 release · Sol/Terra/Luna pricing · TerminalBench 91.9% · Government preview lock · July GA timeline · Six-step runbook
If you are an AI developer, API buyer, or Cursor/Codex user trying to decide whether to re-architect around OpenAI's June 26 launch, the answer is not simple: GPT-5.6 Sol, Terra, and Luna ship with record TerminalBench scores and solar-system naming, but only about 20 vetted partners can access them today while the U.S. government completes its first-ever frontier-model review. This article delivers the verified release facts, Sol/Terra/Luna pricing and modes, benchmark tables versus Claude Mythos 5, safety mechanisms, a July access timeline, use-case recommendations, and a six-step production runbook so you can plan without betting your stack on preview-only access.
Bottom line: OpenAI dropped GPT-5.6 on June 26, 2026 with a new solar naming scheme — Sol (flagship), Terra (balanced), and Luna (lightweight). Sol's Ultra multi-agent mode tops TerminalBench 2.1 at 91.9%, dethroning Claude Mythos 5 after just 17 days at #1. Yet broad ChatGPT and API access remains weeks away, and Polymarket still prices full GA by July 31 at roughly 87%. Teams that rushed on our June leak intelligence now face a different problem: the model exists, but most developers cannot call it.
Partner-only preview: Only ~20 government-approved trusted organizations can reach Sol, Terra, and Luna via API and Codex. General ChatGPT users see nothing yet — weeks until GA.
First U.S. release restriction: President Trump's June 2 executive order triggered a White House request that OpenAI limit rollout. This is the first time Washington formally gated a frontier model — a precedent with export-control echoes for Anthropic's Fable 5 shutdown.
Ultra mode token economics: Sol's multi-agent Ultra mode drives benchmark records but burns significantly more output tokens than standard mode — easy to blow budgets if you route every request through Ultra.
Big Three blocked in June: OpenAI preview-locked GPT-5.6, Anthropic forced Mythos 5 and Fable 5 offline June 12, and Google delayed Gemini 3.5 Pro to July. No Western lab shipped a fully open flagship this month.
Incomplete system card: SWE-Bench Pro and other agentic scores for GPT-5.6 are not fully published. TerminalBench leadership is verified; every other benchmark comparison against Claude remains provisional.
June 2026 was supposed to be the biggest AI release month in history. Instead, all three Western frontier families got stuck at the door — preview lock, export control, or delay.
OpenAI introduced celestial naming for the first time. Sol targets maximum capability with new Max (slow, accurate) and Ultra (multi-agent parallel) reasoning modes. Terra matches GPT-5.5 performance at half Sol's cost. Luna is the budget tier — yet it still earned OpenAI's "High" cybersecurity rating, a first for a non-flagship model in the same family.
| Model | Best For | Input / Output | Context | Highlight |
|---|---|---|---|---|
| GPT-5.6 Sol | Complex coding, security research, long-horizon agents | $5 / $30 per 1M tokens | ~1.5M tokens | Max + Ultra modes; TerminalBench #1 |
| GPT-5.6 Terra | High-volume business docs, support, internal tools | $2.50 / $15 per 1M tokens | ~1.5M tokens | GPT-5.5-level at 50% lower cost |
| GPT-5.6 Luna | Summarization, drafting, routine automation | $1 / $6 per 1M tokens | ~1.5M tokens | 80% cheaper than Sol; High cyber rating |
| Model | Input | Output | Notes |
|---|---|---|---|
| GPT-5.6 Sol | $5/M | $30/M | Same price as GPT-5.5, much higher performance |
| GPT-5.6 Terra | $2.50/M | $15/M | 50% cheaper than Sol; GPT-5.5 parity |
| GPT-5.6 Luna | $1/M | $6/M | 80% cheaper than Sol |
| Claude Fable 5 | $10/M | $50/M | Offline since June 12 export-control order |
GPT-5.6 is the first OpenAI family where all three tiers crossed the internal "High" cybersecurity classification. Benchmark leadership is clearest on agentic coding and security research; life-science scores show meaningful gains over GPT-5.5 as well.
TerminalBench 2.1 runs 89 complex command-line planning challenges — multi-step tool use, iterative repair, and task coordination closer to real agent work than single-shot code completion.
| Model | Score | Mode |
|---|---|---|
| GPT-5.6 Sol | 91.9% | Ultra (multi-agent) |
| GPT-5.6 Sol | 88.8% | Standard |
| Claude Mythos 5 | 88.0% | Standard |
| GPT-5.5 | 83.4% | Standard |
| Gemini 3.1 Pro Preview | 70.7% | Standard |
Mythos 5 had held the top spot for only 17 days since its June 9 coronation before Sol displaced it.
| Model | Task Completion (Code Mode) |
|---|---|
| GPT-5.6 Sol | 50.9% — only model above 50% |
| GPT-5.6 Luna | Slightly above GPT-5.5 |
| Model | CTF Hit Rate |
|---|---|
| Sol | 96.7% |
| Terra | 91.84% |
| Luna | 85.19% |
On ExploitBench, Sol matches Anthropic's Mythos Preview while using roughly one-third the output tokens — comparable vulnerability-research capability at dramatically lower cost.
Safety boundary: OpenAI red-teaming confirmed Sol can identify vulnerabilities and exploit primitives in Chromium and Firefox codebases, but cannot autonomously construct complete, functional exploit chains against hardened targets. It remains below OpenAI's "Cyber Critical" threshold.
On June 2, 2026, President Trump signed an executive order allowing U.S. agencies up to 30 days of pre-release access to review frontier AI models. On June 26, following a White House request coordinated by OSTP and the Office of the National Cyber Director, OpenAI agreed to limit GPT-5.6 to approximately 20 pre-approved trusted partners. This is the first time the U.S. government has formally required an AI company to restrict a model's public release.
OpenAI complied but pushed back publicly: "We don't believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them."
| Company | Model | Status |
|---|---|---|
| OpenAI | GPT-5.6 Sol / Terra / Luna | Limited preview (~20 orgs) |
| Anthropic | Claude Fable 5 / Mythos 5 | Forced offline June 12 (export control) |
| Gemini 3.5 Pro | Delayed to July (originally June) |
| Dimension | GPT-5.6 Sol | Claude Mythos 5 |
|---|---|---|
| TerminalBench 2.1 | 91.9% (Ultra) / 88.8% standard | 88.0% |
| ExploitBench | Near-identical; ~1/3 output tokens | Strong (restricted access) |
| Pricing | $5 / $30 per 1M tokens | $10 / $50 (currently offline) |
| Availability | Preview → GA within weeks | Offline (U.S. export control) |
| Context window | ~1.5M tokens | 200K tokens |
Sol leads on TerminalBench and offers comparable security-research capability at half Fable 5's price. Mythos 5 may still lead on SWE-Bench Pro and other benchmarks until OpenAI publishes the full system card.
Starting July 2026, GPT-5.6 Sol deploys on Cerebras hardware for select enterprise customers at up to 750 tokens per second — roughly 5× to 15× faster than today's 50–150 tok/s frontier models. A 10-second response could drop under one second for real-time coding assistants and live agent UIs.
Access timeline: Right now (~20 partners via API/Codex only). July 2026: ChatGPT GA (Plus/Pro first), public API, and Cerebras-accelerated Sol. Polymarket assigns roughly 87% probability to broad release by July 31.
Do not re-architect production on preview-only access. The runbook separates actions you can take today from post-GA checks once ChatGPT and API endpoints open broadly.
Hold your current stack: Keep GPT-5.5, Claude Opus 4.8, or Sonnet 4.6 in production until Sol/Terra/Luna hit general API availability. Preview scores do not guarantee your workload performance.
Map workloads to tiers now: Route complex agent coding to Sol (Ultra only when justified), high-volume business logic to Terra, and summarization or classification to Luna. Document token budgets before GA spikes costs.
Monitor GA signals: Track openai.com/blog, platform.openai.com/docs, and Polymarket's July 31 contract. Set status-page alerts for API availability the day ChatGPT launches — historically 24–48 hours ahead of API.
Benchmark your own workloads post-GA: Run TerminalBench-style multi-step tasks, frontend generation, and long-context retrieval on Sol standard vs Ultra. Do not assume Ultra's 91.9% translates to your repo structure.
Plan July Cerebras latency tests: If sub-second streaming matters (live coding, customer-facing agents), queue enterprise Cerebras access early — initial capacity is limited.
Maintain multi-vendor fallback: June proved no frontier model is permanently available. Document export-control exposure for foreign staff and keep Anthropic/OpenAI/Gemini routing in your gateway config.
| Your Need | Recommended Model |
|---|---|
| Complex coding agents, multi-step SWE workflows | Sol (Ultra for hardest tasks) |
| Enterprise docs, support tickets, scaled API calls | Terra |
| Summarization, drafting, routine automation | Luna |
| GPT-5.5 performance at half the cost | Terra |
| Latency-critical apps after July | Sol on Cerebras (750 tok/s) |
export PRIMARY_MODEL="gpt-5.5" export PREVIEW_TARGET="gpt-5.6-sol" export FALLBACK_MODELS="claude-opus-4-8,gpt-5.5,gemini/gemini-2.5-pro" curl -s https://status.openai.com/api/v2/status.json | jq '.status.description'
Running Sol Ultra agents on a laptop means Background Agents stop when you close the lid, Linux VPS hosts lack Metal and Keychain boundaries for Codex, and shared dev machines create API key collisions when two agent loops fire at once. Chasing preview-only models on unstable hardware wastes the week between partner access and July GA. For teams that need 24/7 Cloud Agents, persistent Cursor Rules, and lid-closed compile chains while they A/B test Sol, Terra, and Luna the day API opens, a dedicated Mac host beats duct-taping fallbacks across personal hardware. VpsMesh Mac Mini M4 cloud rental delivers launchd reliability, SSH access, and monthly billing in one production node — see rental pricing, the help center for deployment, and the order page to provision before July GA.
Not yet for the general public. As of June 27, 2026, only about 20 vetted partner organizations can access Sol, Terra, and Luna via API and Codex. Full ChatGPT rollout is expected within weeks — Polymarket prices July 31 GA at roughly 87%.
Sol leads on TerminalBench 2.1 at 91.9% (Ultra) versus Claude Mythos 5 at 88%. Claude Fable 5 still leads on SWE-Bench Pro, but official GPT-5.6 SWE-Bench scores have not been published. Sol is the better value — comparable or better agentic coding at roughly half Fable 5's price.
Ultra mode deploys multiple AI subagents that split a complex task, execute in parallel, and synthesize a unified result. It drove Sol's 91.9% TerminalBench record but consumes significantly more tokens than standard mode — use it only for genuinely hard agent workflows.
Following President Trump's June 2, 2026 executive order, the White House requested OpenAI limit GPT-5.6 during a government security review. This is the first time Washington formally required an AI company to restrict a frontier release. OpenAI complied but stated it opposes this becoming permanent practice.
Up to 750 tokens per second for GPT-5.6 Sol on Cerebras starting July 2026 — roughly 5–15× faster than most current frontier models at 50–150 tok/s. Initial access is limited to select enterprise customers.
Reported at approximately 1.5 million tokens across Sol, Terra, and Luna — up from GPT-5.5's 1M. Official confirmation is expected with the full system card at general availability.
Keep production on GPT-5.5 or Claude Opus 4.8 until API GA, but provision a 24/7 Mac host now so you can benchmark Sol/Terra/Luna the day endpoints open. See Mac Mini M4 cloud rental pricing and the help center for deployment steps.