GPT-5.6 Is Here: Sol, Terra, and Luna at a Glance, Pricing, Benchmarks, Availability

OpenAI rolled out GPT-5.6 worldwide on July 9, 2026 after a preview phase with Trusted Partners, offering the model for the first time in three fixed capacity tiers: Sol (flagship), Terra (everyday), and Luna (volume). API prices range from $1 to $30 per million tokens, available in ChatGPT, Codex, and the API.

As of: July 13, 2026

What is new in GPT-5.6?

Until now, a single model stood behind each GPT version number. With GPT-5.6, OpenAI separates generation from capability tier: the version number marks the generation, while Sol, Terra, and Luna, as „durable capability tiers“, can now be developed further independently, without a full version jump. GPT-5.6 followed a limited preview with Trusted Partners and was rolled out globally on July 9, 2026 within roughly 24 hours, available in ChatGPT, Codex, and the OpenAI API. Sol is the flagship for the most demanding reasoning and coding tasks, Terra the balanced everyday option that, according to OpenAI, matches GPT-5.5's performance at half the price, and Luna the fastest, cheapest model for volume workloads such as bulk processing or simple automation.

What do Sol, Terra, and Luna cost?

The three tiers differ significantly in API price per million tokens:

Model	Input	Output
Sol	$5	$30
Terra	$2.50	$15
Luna	$1	$6

New is more predictable prompt caching with explicit cache breakpoints and a minimum cache lifetime of 30 minutes. Cache reads remain attractive at a 90% discount versus the input price, but cache writes now cost 1.25 times the input price: for the first time, OpenAI also charges a fee for writing to the cache. It is worth factoring this into the calculation for applications with recurring, long system prompts.

Who gets which model in ChatGPT and Codex?

Access is tiered by subscription plan and product:

Product / Plan	Access
ChatGPT Plus/Pro/Business/Enterprise	Sol (from „medium effort“ upward); Pro/Enterprise additionally get „Sol Pro“
ChatGPT Work & Codex, Free/Go	Terra
ChatGPT Work & Codex, Plus and above	free choice between Sol, Terra, and Luna, including effort level

New is the effort level „max“ above „xhigh“, available to everyone with GPT-5.6 access in Work and Codex. „Ultra“ coordinates four parallel agents by default and is available in ChatGPT Work for Pro/Enterprise as well as in Codex from Plus upward. In the API, multi-agent orchestration is available as a beta in the Responses API, complemented by „Programmatic Tool Calling“: the model writes and runs small programs that coordinate tool calls and filter intermediate results, which OpenAI says is compatible with Zero Data Retention.

How does GPT-5.6 compare to Claude Fable 5?

OpenAI compares Sol with Anthropic's Claude Fable 5 in its own table, and the picture is mixed. On Agents' Last Exam (55 scoring fields, long professional workflows), Sol reaches 52.7% according to OpenAI, ahead of GPT-5.5 (46.9%) and Fable 5 (40.5%). On the Artificial Analysis Coding Agent Index v1.1, Sol achieves a new best score of 80 points according to OpenAI, ahead of Terra (77.4), Fable 5 (77.2), Luna (74.6), and Opus 4.8 (72.5). On Terminal-Bench 2.1, Sol scores 88.8% according to OpenAI (Ultra 91.9%), narrowly ahead of Claude Mythos 5 (88%) and Fable 5 (83.1%). On OSWorld 2.0 (computer use), Sol reaches 62.6% according to OpenAI, ahead of Opus 4.8 (54.8%); on BrowseComp, Sol scores 90.4% (Ultra 92.2%).

In other disciplines, Fable 5 stays ahead, also according to OpenAI's own table. On the Artificial Analysis Intelligence Index v4.1, Fable 5 leads with 59.9 points, narrowly ahead of Sol (58.9). On GDPval-AA v2, an Elo measure for knowledge work, Fable 5 leads with 1759.6 versus Sol's 1747.8. On SWE-Bench Pro, Fable 5 (80%) and Claude Mythos 5 (80.3%) are clearly ahead of Sol (64.6%). On HealthBench Professional, Fable 5 reaches 60.9% compared to 60.5% for Sol.

The independent measurement from Artificial Analysis confirms the mixed picture: Sol in „max“ mode reaches 59 points on the Intelligence Index, one point behind Fable 5 (60), at roughly a third of the cost per task (Sol $1.04, Terra $0.55, Luna $0.21 per index task). OpenAI puts Sol on the Intelligence Index at roughly 61% faster than Fable 5 at about half the estimated cost, and for coding at less than half as many output tokens. We have put together the detailed tier-by-tier comparison in GPT-5.6 vs. Claude Fable 5 in a tiering comparison.

What do max and ultra mean?

„Max“ is a new effort level above „xhigh“ and is available to all users with GPT-5.6 access in ChatGPT Work and Codex: it increases reasoning effort and therefore tends to improve quality, at the cost of speed and price. „Ultra“ goes further and coordinates four parallel agents on the same task by default, available in ChatGPT Work for Pro and Enterprise accounts as well as in Codex from Plus upward. For businesses, higher effort levels and multi-agent modes can improve result quality on complex tasks, but they also noticeably increase token and cost consumption.

What does this mean for businesses?

For managing directors and IT decision-makers at SMEs, three practical points emerge. First, the tier choice pays off by workload: Sol for complex individual tasks with high reasoning or coding demands, Terra for productive everyday use with a good price-performance ratio, Luna for volume applications such as classification or summarization at scale. Second, price pressure in the frontier market increases buyers' negotiating leverage: OpenAI positions Terra at GPT-5.5's performance level for half the price, while Claude Fable 5 continues to lead in several benchmarks and, for now, remains testable at no extra cost in existing paid Claude subscriptions until July 19, 2026; more on that in our article on free Fable 5 access. Third, model choice therefore belongs structurally in the architecture decision: anyone who wants to flexibly use Sol, Terra, Luna, and alternatives such as Fable 5 should have their AI functions developed so that the underlying model can be swapped per use case, instead of committing early to one provider. That is exactly the approach we take with model-independent AI development.

Frequently asked questions about GPT-5.6

When did GPT-5.6 become available?

OpenAI released GPT-5.6 on July 9, 2026 after a limited preview with Trusted Partners; the global rollout in ChatGPT, Codex, and the API took place within roughly 24 hours.

What distinguishes Sol, Terra, and Luna?

Sol is the flagship model for the most demanding tasks, Terra a balanced everyday model that, according to OpenAI, matches GPT-5.5's performance at half the price, and Luna the fastest and cheapest model for volume workloads.

What does GPT-5.6 cost in the API?

Per million tokens: Sol $5 input / $30 output, Terra $2.50 / $15, Luna $1 / $6. Cache reads remain discounted by 90%, and cache writes now cost 1.25 times the input price with a minimum lifetime of 30 minutes.

Is GPT-5.6 better than Claude Fable 5?

That depends on the discipline. According to OpenAI's own table, Sol leads in agent workflows, coding agent benchmarks, terminal tasks, and computer use, while Fable 5 leads on the Artificial Analysis Intelligence Index, GDPval-AA, SWE-Bench Pro, and HealthBench Professional.

How secure is GPT-5.6 in cybersecurity?

OpenAI describes GPT-5.6 as its strongest cyber model to date (ExploitBench 73.5% versus 47.9% for GPT-5.5), but has also tightened its safeguards at the same time: according to OpenAI, the system blocks about ten times more potentially harmful activity than earlier models, and the most sensitive capabilities remain reserved for verified users in the „Trusted Access for Cyber“ program, which will require hardware passkeys starting September 1. Ahead of release, OpenAI carried out automated red-teaming covering roughly 700,000 A100e GPU hours; according to the company, the models do not cross the „Critical“ threshold of its in-house safety framework in the bio and cyber domains.

Where do I find GPT-5.6 in ChatGPT and Codex?

Plus, Pro, Business, and Enterprise users get Sol from the „medium“ effort level directly in ChatGPT, with Pro/Enterprise additionally getting „Sol Pro“. In ChatGPT Work and Codex, Free and Go users get Terra, while Plus users and above can choose freely between Sol, Terra, and Luna.

Sources

• OpenAI: GPT-5.6 Announcement
• Artificial Analysis: GPT-5.6 has landed

Prices and terms as stated by the provider, as of July 13, 2026, without guarantee. This article does not constitute purchasing advice.