GPT-5.6 Sol, Terra & Luna: Pricing & Access

Introducing the GPT-5.6 family

OpenAI's GPT-5.6 release introduces three distinct model tiers that replace and expand upon the GPT-5.5 lineup. The family is designed to cover the full spectrum of AI workloads — from lightweight chat and classification to frontier agentic reasoning — with clear pricing differentiation between tiers.

Sol is the flagship model, built for long-horizon agentic work, advanced coding, scientific reasoning, and cybersecurity. It introduces a max reasoning effort setting and an ultra mode that uses subagents to parallelize complex tasks. Sol sets a new state of the art on Terminal-Bench 2.1 and ExploitBench, making it the most capable model in the GPT lineup for agent-based coding and security analysis.

Terra is the balanced option, offering GPT-5.5-competitive performance at roughly half the cost. It is suitable for production workloads that need strong general reasoning but do not require frontier capability. Most teams will find Terra is their default model for day-to-day coding, analysis, and automation.

Luna is the entry point to the GPT-5.6 family — the fastest and most affordable tier. It is designed for high-volume, low-latency use cases such as chat, classification, summarization, and any workflow where speed and cost are the primary considerations.

Official pricing

All three models share a 128K max output token limit and support prompt caching at a 90% discount on cached input. Cache writes are charged at 1.25 times the uncached input rate, and batch API calls receive a 50% discount on both input and output pricing.

Source: OpenAI API pricing as of July 9, 2026.

Benchmark highlights

Sol achieves state-of-the-art results on Terminal-Bench 2.1, which evaluates command-line workflows that require planning, iteration, and multi-tool coordination. This benchmark is directly relevant to coding agents like Codex and Cline, because it tests the same capabilities those tools need — the ability to understand a goal, break it into steps, execute commands, interpret results, and adjust course.

On ExploitBench, Sol delivers competitive cybersecurity results while using roughly one-third the output tokens of other frontier systems. This efficiency means lower latency and lower cost per security analysis task. In scientific reasoning, Sol scores 53.5% on the Virology Capabilities Test, 60.0% on Molecular Biology, and 68.3% on World-Class Bio — approximately 9 points above GPT-5.5 — making it a strong tool for research and analysis workflows.

What these benchmarks reveal is that Sol is not just more powerful than previous models — it is also more efficient. It produces better results with fewer tokens, which directly translates to lower cost per task when accessed through a token-based API.

What GPT-5.6 means for API users

The three-tier pricing structure — Luna, Terra, Sol — gives API users a clear upgrade path. Start with Luna for high-volume, low-cost work. Use Terra as your default production model for most tasks. Reserve Sol for the most complex agentic and reasoning work. This tiered approach mirrors what token-based API platforms offer: the flexibility to match model capability to task difficulty, and pay only for what you need.

Sol's improved tool-use performance and ultra mode make it especially relevant for agent frameworks. OpenClaw, Claude Code, Cline, Codex, and custom orchestrators all benefit from stronger multi-step reasoning. And because Sol is more token-efficient than its predecessors, it delivers more value per token spent.

APItokendeal provides OpenAI-compatible access to a wide range of models including GPT-5.4, GPT-5.5, Claude, DeepSeek, Gemini, and more through a single API key. As GPT-5.6 models become available through upstream providers, they will be added to the platform with clear weightage pricing so you know exactly what each call costs.