GPT-5.6 Sol, Terra, Luna: Skills Setup for Codex CLI
GPT-5.6 launched in three tiers. How to use SKILL.md skills with Sol, Terra, and Luna in Codex CLI.
Quick Answer: GPT-5.6 launched June 26, 2026 in three tiers: Sol (flagship), Terra (GPT-5.5 level at half price), and Luna (cheapest). Currently limited preview through API and Codex only, about 20 approved partners. SKILL.md skills work with all three tiers. Broader access is expected within weeks.
OpenAI shipped GPT-5.6 on June 26, 2026. This is the first time OpenAI has launched a model family with three distinct tiers instead of a single flagship. Sol is the frontier model for complex coding and agentic workflows. Terra matches GPT-5.5 performance at roughly half the cost. Luna is the fast, cheap option for high-volume work.
The launch came with an unusual constraint: the U.S. government asked OpenAI to limit the initial rollout to about 20 pre-approved partner organizations. GPT-5.6 is not available in ChatGPT during the preview period. Broader access is planned in the coming weeks.
The three tiers explained
Sol is the flagship. It scored 91.9% on Terminal-Bench 2.1 in Ultra mode (which uses subagents), surpassing Claude Mythos 5. It introduces a max reasoning effort setting and is built for complex coding, cybersecurity, and long-horizon agentic work. Pricing: $5 per million input tokens, $30 per million output tokens.
Terra is the practical choice for most developers. OpenAI positions it at GPT-5.5-level performance at roughly half the cost. Pricing: $2.50/$15 per million tokens. For everyday coding tasks, tests, and code review, Terra is likely the sweet spot once it becomes available.
Luna is the volume tier. Fastest response times, lowest cost ($1/$6 per million tokens), designed for tasks where speed matters more than peak reasoning. Good for high-volume automation, summarization, and simple code generation.
Recommended skills
Custom Data RAG Chatbot Builder
by tudor.ai
Build a full-stack AI chatbot trained on your own documents across any industry…
WooCommerce Crypto Payments — Accept BTC, ETH, USDC & 100+ Tokens
by zoninmane
Add non-custodial, zero-KYC crypto payments to WooCommerce for instant wallet se…
code-repair-spine
by Corey Jacobs
Generate source-safe repository audits and repair handoff bundles without mutati…
Using SKILL.md skills with GPT-5.6
SKILL.md skills work with Codex CLI regardless of which GPT model version it runs. The skills define what the agent should do. The model determines how well it does it. A code review skill or testing skill installed in your .skills/ directory will work with Sol, Terra, or Luna once you have access.
To install skills for Codex CLI:
- Browse skills at agensi.io/skills
- Download and unzip to your Codex skills directory
- Skills activate automatically on the next session
For a full setup guide, see how to install skills in Codex CLI.
What this means for developers
If you are currently using GPT-5.5 through Codex CLI, the upgrade path is straightforward once access opens. Terra gives you the same performance at half the cost. Sol gives you frontier-level capability for complex tasks.
The tiered pricing model means you can route different tasks to different models. Use Luna for simple boilerplate generation, Terra for everyday development, and Sol for complex refactors and architecture decisions. Your SKILL.md skills work across all three tiers.
For developers not yet in the preview: GPT-5.5 remains fully capable and available. The skills ecosystem works the same regardless of model version. Browse 1,600+ skills on the Agensi marketplace.
GPT-5.6 vs Claude Fable 5 vs Claude Opus 4.7
Sol Ultra leads Terminal-Bench 2.1 at 91.9%. Fable 5 leads SWE-Bench Verified at 88.6%. Opus 4.7 remains the most widely available frontier coding model at $15/$75 per million tokens through Claude Code ($20/month Pro).
Fable 5 availability is currently limited following export control restrictions in mid-June 2026. Opus 4.7 remains the most accessible frontier option through Claude Code. GPT-5.6 Sol will likely become the primary Codex CLI model once the preview period ends.
For a full comparison of all coding agents and their model support, see best AI coding assistants 2026.