GPT-5.6 Lands Tomorrow: OpenAI’s Most Capable Model Family Yet
OpenAI is releasing GPT-5.6 tomorrow, Thursday, July 9, 2026, moving from the late-June limited preview to general availability for the full family: flagship Sol, balanced Terra, and fast, affordable Luna. The launch brings a new naming system, deeper reasoning controls, state-of-the-art coding results, stronger biology evaluations, and OpenAI’s most robust safety stack to date.
Three models, one generation
GPT-5.6 is a family rather than a single model. Sol is the flagship, designed for the hardest reasoning and agentic work. Terra is positioned as a balanced everyday model that delivers competitive performance to GPT-5.5 at roughly half the price. Luna is the fast, most affordable tier, aimed at high-volume, cost-sensitive workloads.
The naming itself is new. Going forward, the number marks the generation, while Sol, Terra, and Luna are durable capability tiers that can advance on their own schedule, so a future Luna or Terra upgrade does not have to wait for the next generation drop.
What new reasoning controls are in GPT-5.6?
Two new dials are landing alongside the models. A new max reasoning effort gives Sol the most time to reason deeply on a single task. A new ultra mode goes further, dispatching subagents to break complex work into parallel pieces and accelerate it. Together, they give developers finer control over how much time and compute the model spends before returning an answer.
Coding: state of the art on Terminal-Bench 2.1
GPT-5.6 Sol sets a new state of the art on Terminal-Bench 2.1, a benchmark for command-line workflows that require planning, iteration, and tool coordination. That matters because real-world agentic coding rarely looks like a single prompt and response. It looks like a model reading files, running commands, inspecting errors, and looping until a task is done. Terminal-Bench 2.1 tests exactly that loop.
Biology: stronger, fewer tokens
On GeneBench v1, which evaluates long-horizon genomics and quantitative-biology analyses, GPT-5.6 Sol posts stronger results than GPT-5.5 while using fewer tokens. For labs running long analytical pipelines, that combination is the one that actually moves budgets and unlocks new experiments.
Cybersecurity: most capable yet, with safeguards that scale
Sol is also OpenAI’s most capable cybersecurity model. On ExploitBench it is competitive with Mythos Preview while using only about one third of the output tokens. On ExploitGym, a benchmark created by UC Berkeley researchers in collaboration with OpenAI and other frontier labs, Sol, Terra, and Luna all show strong improvements as reasoning effort increases, a sign that the gains come from the model thinking more, not from raw size.
OpenAI is pairing those gains with safeguards scaled to match. In Chromium and Firefox evaluations, Sol identified bugs and exploitation primitives but did not autonomously produce a functional full-chain exploit under the conditions tested, and the company states it does not cross the Cyber Critical threshold under its Preparedness Framework. The framing: better at helping defenders find and fix vulnerabilities than at reliably running end-to-end attacks.
The safety stack behind the launch
GPT-5.6 ships with what OpenAI calls its most robust safeguard stack to date, with configurations matched to each tier:
- Refusal training covers prohibited cyber assistance, including disguised-intent and jailbreak attempts.
- Real-time cyber and biology misuse classifiers evaluate output as it is generated. For higher-risk cases, generation can pause while a larger reasoning model reviews the conversation.
- Flagged activity can trigger account-level review, with differentiated access for users who repeatedly cross the line.
- Automated red-teaming ran for more than 700,000 A100-equivalent GPU hours, searching for universal jailbreaks that work across many prompts or contexts.
- Third-party human red-teaming continues through the preview period, paired with a rapid-response process to reproduce, prioritize, and remediate new jailbreaks.
OpenAI also notes that during the preview, some legitimate requests may be blocked or refused, and that feedback from the preview period will be used to reduce unnecessary blocks before wider release.
The preview process, and a note on its future
The preview began with a small group of trusted partners, in coordination with the U.S. government, as OpenAI worked on the cyber Executive Order framework. OpenAI states plainly that it does not want this kind of pre-release government access process to become the long-term default, framing it as a short-term step toward broader availability while a repeatable process for future releases is developed.
Pricing and prompt caching
GPT-5.6 is priced per 1M tokens across the three tiers:
- Sol: $5 input / $30 output
- Terra: $2.50 input / $15 output
- Luna: $1 input / $6 output
Prompt caching is getting more predictable. The new system supports explicit cache breakpoints and a 30-minute minimum cache life. Cache writes are billed at 1.25x the uncached input rate, and cache reads continue to receive the 90% cached-input discount, so long-running agents and chat sessions can plan their caching strategy in advance.
Cerebras partnership brings 750 tokens per second
In July, GPT-5.6 Sol is also coming to Cerebras at up to 750 tokens per second for select customers. For teams that have been waiting for frontier intelligence at inference speeds closer to in-product responsiveness, that is the headline infrastructure beat of the launch.
What to expect on day one
Tomorrow is the general release. ChatGPT, Codex, and the API should see broader access roll out as OpenAI moves past the partner-only preview, with the Cerebras-hosted Sol available to a narrower initial customer set. Developers shipping agentic coding workflows, biology pipelines, and defensive security tools have the most to test first: max and ultra reasoning for hard tasks, the new caching rules for cost control, and the stronger safety stack that may occasionally block requests and learn from feedback over the coming weeks.
FAQ
When is GPT-5.6 releasing and which models are included?
GPT-5.6 releases tomorrow, Thursday, July 9, 2026, moving from the late-June limited preview to general availability for the full family: flagship Sol, balanced Terra, and fast, affordable Luna.
What are the new reasoning controls in GPT-5.6?
GPT-5.6 introduces two new dials: a max reasoning effort that gives Sol the most time to reason deeply on a single task, and an ultra mode that dispatches subagents to break complex work into parallel pieces and accelerate it.
How much does GPT-5.6 cost per million tokens?
Pricing per 1M tokens is Sol at $5 input / $30 output, Terra at $2.50 input / $15 output, and Luna at $1 input / $6 output, with prompt caching supporting explicit breakpoints, a 30-minute minimum cache life, 1.25x cache writes, and a 90% cached-input discount on reads.