GPT-5.6 Released: How OpenAI Reframes Model Capabilities from Sol to Luna

OpenAI officially released the GPT-5.6 model family on July 9, 2026. Instead of focusing on a single “most powerful model,” the release separates capability and cost into three durable tiers: flagship GPT-5.6 Sol, balanced GPT-5.6 Terra, and value-oriented GPT-5.6 Luna.

The goal is to make model selection fit real workloads: use Sol when complex tasks demand the highest completion quality; choose Terra for everyday office work and coding; and hand classification, extraction, and batch processing to Luna.

Positioning of the three tiers

OpenAI says the number represents the model generation, while Sol, Terra, and Luna are capability tiers that can evolve independently. Future upgrades therefore do not need to arrive across the entire GPT-5.6 line at once; each tier can improve on its own schedule.

Sol is about more than benchmarks

GPT-5.6 Sol targets tasks that require sustained planning, tool use, and result checking. OpenAI says it reaches new frontier levels in coding, knowledge work, cybersecurity, and scientific research, completing some tasks with fewer tokens and less time.

Sol also improves computer operation and design judgment. It can inspect rendered pages, identify layout or interaction problems, and keep editing; in presentations, documents, and spreadsheets, it more reliably follows reference fonts, spacing, colors, and page structures.

For developers, Sol is especially suitable for:

• Modifying large codebases across multiple modules, running tests, and diagnosing failures on its own.
• Turning documents, online sources, and tool results into professional deliverables.
• Running agents that need long context and multiple rounds of decisions.
• Tasks where a single failure is expensive and human rework must be minimized.

How Terra and Luna reduce cost

Terra is the best default tier for many workloads. Routine content generation, ordinary code changes, meeting summaries, document Q&A, and standard tool calls usually do not justify Sol’s price.

Luna fits high-throughput tasks whose results are easy to verify. For example, use Luna for text classification and field extraction first, then send anomalous samples to Terra or Sol for review. This routing controls the budget better than always using the flagship model and lets the most capable model focus on genuinely difficult cases.

A practical tiered workflow is:

1. Luna handles preprocessing, cleaning, classification, and simple summaries.
2. Terra handles default conversations, routine coding, and most business requests.
3. Escalate low-confidence, failed-validation, or repeatedly retried tasks to Sol.
4. Track quality, latency, and token cost for each tier, then adjust routing ratios.

What are max and ultra?

max is not a new model but a higher reasoning intensity. It gives GPT-5.6 more time than xhigh to explore approaches, run checks, and revise results. It suits difficult problems, but increases computation and waiting time.

ultra coordinates multiple agents working in parallel. It defaults to four agents handling different workflows and then aggregates their results. It is useful for research, engineering, and analysis that can be split into independent branches. For simple questions, ultra usually only adds cost.

In the API, developers can use the Responses API multi-agent beta to build parallel workflows similar to ultra. Programmatic Tool Calling also lets a model run lightweight programs in memory, filter large tool responses first, and return only useful results to context, reducing model round trips and token usage.

Availability in ChatGPT, Codex, and the API

GPT-5.6 is rolling out globally. OpenAI says the rollout will continue during the first 24 hours after release:

• ChatGPT Plus, Pro, Business, and Enterprise users can use Sol; Pro and Enterprise can also select Sol Pro.
• In ChatGPT Work and Codex, Free and Go users can use Terra; Plus, Pro, Business, and Enterprise users can choose Sol, Terra, or Luna.
• max is available to ChatGPT Work and Codex users who have GPT-5.6 access.
• ultra is available in ChatGPT Work for Pro and Enterprise, and in Codex for Plus and higher tiers.
• API developers can call Sol, Terra, and Luna and use Programmatic Tool Calling and the multi-agent beta.

If the model options are not visible yet, the staged rollout may simply be incomplete rather than unsupported by the account.

Caching rules have changed too

GPT-5.6 supports explicit cache breakpoints and a 30-minute minimum cache lifetime. For GPT-5.6 and later models, cache writes cost 1.25 times the uncached input price, while cache reads retain a 90% discount on cached input.

API applications can place stable system prompts, tool definitions, and background material in the stable cache, with live user input appended afterward. This improves cache-hit rates and makes the real cost of each model tier easier to estimate.

GPT-5.6’s safety-first release

As capabilities in cybersecurity and life sciences improve, OpenAI is strengthening its safety layers. GPT-5.6 combines safety behaviors trained into the model, real-time checks, continuous monitoring, and account-level risk controls, with a Trusted Access program for verified defensive cybersecurity work.

OpenAI also acknowledges that stricter safeguards can mistakenly block some legitimate requests. For high-risk capabilities, access is differentiated by user identity, task context, and risk signals rather than exposing the same capability to everyone.

Conclusion: choose a tier first, then increase reasoning

GPT-5.6 offers two levels of choice: first select the cost and capability tier—Sol, Terra, or Luna—then choose standard reasoning, max, or ultra according to task difficulty.

• Highest quality, complex coding, and long-running agents: Sol.
• A balance of quality, speed, and cost: Terra.
• Simple, high-volume, easily checked tasks: Luna.
• A single difficult task: increase reasoning intensity and use max when needed.
• Work that can be split into parallel branches: consider ultra.

Original: GPT-5.6: Frontier intelligence that scales with your ambition