ChatGPT 5.6 (officially GPT-5.6) went generally available on July 9, 2026. It's a family of three models: Sol (flagship), Terra (balanced, roughly GPT-5.5 quality at half the cost), and Luna (fast and cheap). In the regular ChatGPT chat you need at least a Plus plan to use Sol; free users get Terra only inside the new ChatGPT Work and Codex surfaces. API pricing per 1M tokens: Sol $5/$30, Terra $2.50/$15, Luna $1/$6. The headline extras are a new max reasoning effort, an ultra multi-agent mode, and Programmatic Tool Calling for developers.
Searches for ChatGPT 5.6 exploded this week, and for good reason: OpenAI moved GPT-5.6 from a limited preview to general availability on July 9, 2026, rolling it out across ChatGPT, Codex, and the API over about 24 hours. It's also the first release under OpenAI's new naming system, which is exactly why so many people are googling "chatgpt 5.6 sol" and "chatgpt 5.6 release date" right now. Here's what actually shipped, what the three names mean, which plan gets which model, and where it stands against the competition, based on OpenAI's own announcement and eval tables.
ChatGPT 5.6 at a glance
| Sol | Terra | Luna | |
|---|---|---|---|
| Positioning | Flagship, deepest reasoning | Balanced, everyday work | Fast, most affordable |
| API price (per 1M tokens in/out) | $5 / $30 | $2.50 / $15 | $1 / $6 |
| API model ID | gpt-5.6-sol (also the gpt-5.6 alias) |
gpt-5.6-terra |
gpt-5.6-luna |
| In regular ChatGPT chat | Plus and up | — | — |
| In ChatGPT Work / Codex | Plus and up | All plans, including Free | Plus and up |
| Released | Preview June 26 → GA July 9, 2026 | Same | Same |
What "Sol, Terra, Luna" actually means
GPT-5.6 introduces a new naming system. The number (5.6) identifies the model generation, while Sol, Terra, and Luna are durable capability tiers that OpenAI says can advance on their own schedules. In practice it maps closely to what other labs already do: a flagship model (Sol), a mid-tier workhorse (Terra), and a small, fast model for high-volume tasks (Luna).
OpenAI's own framing from the preview announcement: Terra offers performance competitive with GPT-5.5 while being 2x cheaper, and Luna brings strong capability at the lowest cost. Sol is the flagship, and the one OpenAI calls its strongest model yet, with the biggest gains in agentic coding, computer use, and long-horizon tasks.
One naming wrinkle worth knowing: "GPT-5.6 Sol Pro," the option Pro and Enterprise subscribers see in the ChatGPT picker, isn't a separate model. It's the same Sol model running with a "pro" reasoning mode that developers can also enable through the API.
Release date and the unusual two-step launch
GPT-5.6 had a phased release, which is part of why the search interest spiked twice:
The limited preview began June 26, 2026, restricted to a small group of vetted partners through the API and Codex only. OpenAI said it took this step at the request of the U.S. government because of the model's significantly stronger cybersecurity capabilities, and paired the model with its most extensive safety stack to date, including real-time misuse classifiers and over 700,000 GPU-hours of automated red-teaming.
General availability came July 9, 2026, across ChatGPT, Codex, and the API, rolling out globally over roughly 24 hours. Pricing and benchmarks were unchanged from the preview; the news was access. The same morning, OpenAI also launched ChatGPT Work (a new agentic work surface) and merged Codex into the ChatGPT desktop app. OpenAI also said GPT-5.4 will retire on July 23, while GPT-5.5 remains available.
Which ChatGPT plan gets which model
This is the question most people actually need answered, and the answer depends on where in ChatGPT you are:
Regular chat: Plus, Pro, Business, and Enterprise users get GPT-5.6 Sol at medium and higher effort settings. Pro and Enterprise users additionally get Sol Pro for the hardest tasks. OpenAI's access notes list no free-tier option for plain chat.
ChatGPT Work and Codex: Free and Go users get Terra, which makes these surfaces the only no-cost path to GPT-5.6 right now. Paid plans from Plus up can pick between Sol, Terra, and Luna and set a reasoning effort per model.
The max effort and ultra mode: GPT-5.6 adds a max reasoning effort above the old xhigh, available as a settings toggle to everyone with GPT-5.6 access in Work and Codex. ultra is more exotic: it coordinates four agents in parallel (more on some benchmark runs) to finish complex work faster. In ChatGPT Work, ultra is reserved for Pro and Enterprise; in Codex it starts at Plus. Note that ultra burns significantly more tokens by design; it's a spend-more-to-finish-faster lever, not an efficiency feature.
If you're weighing whether a paid ChatGPT plan is worth it against the alternatives, our ChatGPT vs Claude vs Gemini comparison covers the subscription math in detail.
What's new for developers
Three API-side changes stand out beyond the model IDs:
Programmatic Tool Calling (PTC). Instead of round-tripping every tool call through the model, GPT-5.6 can write and run a small program in a hosted runtime that coordinates tools itself: parallel calls, loops, filtering large intermediate outputs. OpenAI's launch post cites customers cutting prompt tokens by 38% (Clio) and total tokens by 63.5% (PlayCo) after adopting the pattern. The honest caveat: those savings came from re-architecting around PTC, not from just swapping the model name.
A multi-agent beta in the Responses API lets GPT-5.6 run concurrent subagents and synthesize their work in a single request, the API equivalent of ChatGPT's ultra mode.
More predictable prompt caching, with explicit cache breakpoints and a 30-minute minimum cache life. Cache writes now bill at 1.25x the input rate; reads keep the 90% discount.
For how this fits into the coding-agent race, see our comparisons of OpenAI Codex vs Claude Code and Cursor vs Copilot vs Claude Code.
How good is it, really?
Going by OpenAI's own published eval tables (no independent audits existed at GA), the picture is strong but split:
Where Sol leads: long-horizon agentic work (Agents' Last Exam: 52.7% vs Claude Opus 4.8's 45.2%), Terminal-Bench 2.1 (88.8%, with the Sol Ultra configuration at 91.9%), and computer use (OSWorld 2.0: 62.6% vs 54.8%), reportedly with far fewer output tokens than competing models.
Where Claude still leads on the same tables: repository-scale software engineering (SWE-Bench Pro: Claude Mythos 5 at 80.3% vs Sol's 64.6%), GDPval's professional-work Elo ranking, and the Artificial Analysis Intelligence Index.
Benchmarks are vendor-reported
Every number above comes from OpenAI's launch materials. Vendor evals are directionally useful but always flatter the vendor's own model. If a specific workload matters to you, test it yourself before switching; both companies offer free tiers or trials that make a quick head-to-head cheap.
The fair one-line summary: GPT-5.6 Sol is a real step up for agentic tasks and computer use, while frontier Claude models keep the lead in heavyweight software engineering. Nobody swept the board. For the wider landscape this release lands in, our latest AI tools roundup has the running scorecard.
Should you upgrade or switch?
If you're a free ChatGPT user, nothing changes in regular chat, but you can try GPT-5.6 Terra today for free inside ChatGPT Work or Codex, which is a genuinely capable model at the price of zero.
If you're on Plus, you already have Sol in chat; the practical upgrade is toggling higher effort settings for hard tasks rather than paying more.
If you're a developer, OpenAI's own migration guidance is to treat this as a tuning pass: test your prompts at the current effort level and one lower, point high-volume classification at Luna ($1/$6 makes it very cheap), and only adopt ultra or PTC where the workload actually decomposes into parallel or predictable steps.
Frequently Asked Questions
Is ChatGPT 5.6 free to use?
Not in the regular chat interface. Free and Go users get GPT-5.6 Terra inside ChatGPT Work and Codex only. In plain chat, GPT-5.6 Sol starts with the Plus plan; Pro and Enterprise add Sol Pro.
What is ChatGPT 5.6 Sol?
Sol is the flagship tier of the GPT-5.6 family, OpenAI's most capable model as of July 2026. In the new naming system the number is the generation and Sol/Terra/Luna are capability tiers, so future releases will keep these names.
When was ChatGPT 5.6 released?
Limited preview on June 26, 2026 for vetted partners via the API and Codex; general availability across ChatGPT, Codex, and the API on July 9, 2026, rolling out over about 24 hours.
How much does the GPT-5.6 API cost?
Per 1M tokens: Sol $5 input / $30 output, Terra $2.50 / $15, Luna $1 / $6. Cache reads get a 90% discount; cache writes bill at 1.25x the input rate.
Is GPT-5.6 better than Claude?
It depends on the task. OpenAI's own tables show Sol ahead on agentic and computer-use benchmarks, while Claude models lead on SWE-Bench Pro and the Intelligence Index. Run your own workload before deciding.
What is ultra mode in GPT-5.6?
A setting that runs multiple agents (four by default) in parallel on one task and synthesizes their work, trading higher token use for faster, stronger results. It's available in ChatGPT Work (Pro/Enterprise) and Codex (Plus and up).
Conclusion
ChatGPT 5.6 is less a single new model than a reorganization of OpenAI's lineup into three clear tiers plus a set of agentic power tools: max effort, ultra mode, and Programmatic Tool Calling. The cheapest way to feel the difference today is free Terra access in ChatGPT Work; the most consequential change for developers is PTC. As always with launch-week claims, trust the direction more than the decimal points, and check the numbers against your own use before moving anything important.
Want to see how the whole field stacks up right now? Start with our up-to-date ChatGPT vs Claude vs Gemini comparison.