Update, July 10, 2026: GPT-5.6 is now generally available. After a limited government-gated preview in late June, OpenAI released GPT-5.6 to everyone on July 9, 2026 across ChatGPT, Codex, and the API. This post has been updated from our preview coverage to reflect the public launch.
GPT-5.6 is OpenAI’s newest model family, and as of July 9, 2026 it is generally available across ChatGPT, Codex, and the API. It is not a single model but three tiers, named Sol, Terra, and Luna, and it introduces two new reasoning modes, max and ultra, along with the strongest agentic-coding results OpenAI has posted yet. It also underpins two products launching alongside it: the new ChatGPT Work agent and the ChatGPT Live voice experience. This piece explains what GPT-5.6 is, how the tiers and reasoning modes work, how it performs, what it costs, and how it compares.
The short version: GPT-5.6 keeps the GPT-5 line’s architecture and pushes hard on long-horizon agentic work, coding, cybersecurity, and biology, while getting more efficient. It splits into a flagship (Sol), a balanced everyday tier (Terra), and a fast, cheap tier (Luna), and adds an "ultra" mode that coordinates subagents in parallel. After an unusual, government-requested delay tied to its cybersecurity capability, it is now available to everyone. For the model it succeeds, see our coverage of GPT-5.5.
What GPT-5.6 is
GPT-5.6 is a model line, not a single model, with three members at launch:
- GPT-5.6 Sol is the flagship, the most capable tier for complex reasoning, long-horizon agentic work, coding, cybersecurity, and biology. It is the tier the new max and ultra reasoning modes attach to, and the one to reach for when correctness matters more than cost.
- GPT-5.6 Terra is the balanced tier for everyday production work. OpenAI positions it as competitive with GPT-5.5 at roughly half the cost.
- GPT-5.6 Luna is the fast, low-cost tier for high-volume workloads where latency and price matter more than the absolute capability ceiling.
The naming is a deliberate change worth understanding. Under the new scheme the generation number and the tier name are independent: the number (5.6) marks the generation, while Sol, Terra, and Luna are durable tier names meant to persist across future generations. So a later GPT-5.7 Sol could ship without renaming the whole family. The practical takeaway is to learn the three tier names, because model selection now happens along that Sol/Terra/Luna axis.
How it works: the max and ultra reasoning modes
The most genuinely new capability is a pair of higher reasoning-effort settings on Sol.
Max gives Sol the most time to reason deeply in a single pass. It is the "think harder, take longer" dial for problems where one deep deliberation beats a faster answer, an extension of the reasoning-effort controls that have existed since the GPT-5 family, pushed to a higher ceiling.
Ultra is the more architecturally interesting one. Instead of a single agent reasoning longer, ultra mode coordinates subagents to work on a complex task in parallel and then integrates the results. In other words, ultra goes beyond what one model instance can do alone by orchestrating several. This is OpenAI productizing the multi-agent pattern that developers previously had to assemble themselves with frameworks and orchestration code. For long, decomposable tasks, a large refactor, a multi-file investigation, a research synthesis, the subagent approach can change the shape of what a single request accomplishes.
The caveat is cost: ultra is the most expensive way to run the model, because subagents multiply token spend. It is a mode to reach for on genuinely hard, parallelizable work, not a default.
How it performs
The performance story centers on agentic, long-horizon work rather than chat quality, and the theme is doing more with fewer tokens.
On Terminal-Bench 2.1, which tests command-line workflows requiring planning, iteration, and tool coordination, GPT-5.6 Sol in ultra mode scores around 91.9%, with Sol on its own near 88.8%, edging Claude’s Mythos 5 line near 88% and clearly ahead of GPT-5.5 near 83%. OpenAI also reports Sol is markedly more token efficient than GPT-5.5 on agentic coding, and describes it as its most capable model yet for cybersecurity work such as vulnerability research, with broad gains in biology as well.
Two honest caveats. These gains are concentrated in agentic and security-adjacent work, so they say more about software engineering and long-horizon tasks than about every kind of query. And benchmarks are a guide, not a verdict; the real test is your own workloads. But the direction is clear: GPT-5.6 is at or near the top on hard coding, and notably more efficient while it gets there.
What it costs
API pricing follows the three tiers, per million tokens:
- Sol: $5 input / $30 output
- Terra: $2.50 input / $15 output
- Luna: $1 input / $6 output
The structure rewards matching the tier to the task: route high-volume, simple work to Luna, run everyday production on Terra, and reserve Sol, especially in max or ultra mode, for the hardest problems where its capability and efficiency justify the higher rate. Because output tokens cost several times more than input across every tier, the shape of your workload drives the bill as much as the tier you pick.
From gated preview to public launch
GPT-5.6’s rollout was unusual. It first appeared in late June as a limited preview restricted to a small set of government-approved partners, after the White House asked OpenAI to stagger access while federal agencies ran a cybersecurity evaluation, citing Sol’s offensive-security capability. OpenAI complied while publicly arguing that such gating should not become the default, since restrictions also keep tools from the defenders and enterprises who need them.
That review has since concluded, and GPT-5.6 is now generally available to everyone through ChatGPT, Codex, and the API. The episode is worth remembering as a sign of how governments may treat frontier models with strong cyber capabilities, but the operative fact today is simple: you can use GPT-5.6 now.
What it powers: ChatGPT Work and ChatGPT Live
GPT-5.6 is also the engine behind two products OpenAI shipped at the same time. ChatGPT Work is a new autonomous agent that carries out complete assignments across your connected apps and files, producing finished documents, spreadsheets, and reports, and running standing automations in the background. ChatGPT Live is the new full-duplex voice experience. Both lean on GPT-5.6’s agentic strength and its ability to delegate hard reasoning, which is a good illustration of the point of this release: the model is built to power agents that do multi-step work, not just answer questions.
How it compares
Against GPT-5.5, GPT-5.6 is an incremental generation focused on agentic coding, security, and the new reasoning modes, with better efficiency rather than a new architecture. Against the leading Claude models like Opus 4.8 and the Fable 5 tier, GPT-5.6 Sol is competitive-to-leading on agentic coding as measured by these benchmarks, with ultra mode opening a wider margin, though one benchmark suite is never the whole story and rivals have their own strengths. The meaningful change from a month ago is availability: where GPT-5.6 was gated behind a government review, it now sits alongside GPT-5.5, Claude, and Gemini as an option you can actually deploy.
The bottom line: GPT-5.6 is OpenAI’s new Sol/Terra/Luna family, now generally available, built for long-horizon agentic work and priced by tier so you can match capability to cost. Its headline additions are the ultra subagent mode and a real jump in token efficiency, and it is the model now powering ChatGPT Work and ChatGPT Live.
Frequently Asked Questions
What is GPT-5.6?
GPT-5.6 is OpenAI’s newest model family, generally available since July 9, 2026 across ChatGPT, Codex, and the API. It ships as three tiers, Sol (flagship), Terra (balanced), and Luna (fast and cheap), and adds two new reasoning modes, max and ultra. It is focused on long-horizon agentic work, coding, cybersecurity, and biology, and it powers the new ChatGPT Work and ChatGPT Live products.
What do Sol, Terra, and Luna mean?
They are durable capability tiers named after the Sun, Earth, and Moon. Sol is the high-intelligence flagship, Terra is the balanced everyday model, and Luna is the fast, low-cost option. The generation number (5.6) is separate from the tier name, so the names are meant to persist across future generations while the number ticks up.
What are the max and ultra reasoning modes?
Max gives the model the most time to reason deeply in a single pass. Ultra goes further by coordinating subagents to work on a complex task in parallel and integrate the results, which is OpenAI productizing the multi-agent pattern. Ultra is the most capable and the most expensive option, because subagents multiply token usage, so it suits genuinely hard, parallelizable work.
How much does GPT-5.6 cost?
API pricing per million tokens is $5 input and $30 output for Sol, $2.50 and $15 for Terra, and $1 and $6 for Luna. Because output costs several times more than input on every tier, the shape of your workload matters as much as which tier you choose. Matching the tier to the task is the main way to control cost.
Can I use GPT-5.6 now?
Yes. After a limited, government-gated preview in late June tied to a federal cybersecurity review, that review concluded and GPT-5.6 became generally available on July 9, 2026 through ChatGPT, Codex, and the API. It now sits alongside GPT-5.5, Claude, and Gemini as an option you can deploy.
How does GPT-5.6 perform on coding?
Strongly. GPT-5.6 Sol posts around 88.8% on Terminal-Bench 2.1, and about 91.9% in ultra mode, edging Claude’s Mythos 5 line near 88% and clearly ahead of GPT-5.5 near 83%, and OpenAI reports it is markedly more token efficient than GPT-5.5 on agentic coding. The gains are concentrated in agentic, long-horizon coding, so results on other kinds of tasks will vary.
What is the difference between ChatGPT 5.6 and GPT-5.6?
GPT-5.6 is the underlying model family. ChatGPT is the consumer product that runs OpenAI’s models. People often say “ChatGPT 5.6” to mean the GPT-5.6 models as they appear inside ChatGPT. Now that GPT-5.6 is generally available, those models are accessible in ChatGPT as well as through Codex and the API.
What does GPT-5.6 power beyond chat?
GPT-5.6 is the engine behind OpenAI’s new ChatGPT Work, an autonomous agent that completes assignments across your apps and files, and ChatGPT Live, the full-duplex voice experience. Both rely on the model’s agentic strength, which reflects the release’s emphasis on models that do multi-step work rather than only answering questions.
