GPT-5.6 Pricing Explained: Sol, Terra, Luna, and Prompt Caching

GPT-5.6 Pricing Explained

OpenAI lists GPT-5.6 pricing across three model sizes: GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna. Sol is the flagship and most capable model. Terra is a strong lower-cost option. Luna is the fastest and most cost-efficient model in the family. During the limited preview, access is available through the OpenAI API and Codex to selected trusted partners and organizations; GPT-5.6 is not available in ChatGPT during the preview.

The official GPT-5.6 prices are listed per 1 million tokens:

GPT-5.6 Sol: $5.00 input, $30.00 output

GPT-5.6 Terra: $2.50 input, $15.00 output

GPT-5.6 Luna: $1.00 input, $6.00 output

OpenAI also says GPT-5.6 introduces more predictable prompt caching, including explicit cache breakpoints and a 30-minute minimum cache life. For GPT-5.6 and later models, cache writes are billed at 1.25 times the model’s uncached input rate, while cache reads continue to receive a 90% cached-input discount.

For developers, product teams, and creator tools, these pricing details matter because GPT-5.6 is not one fixed-cost model. The cost depends on which model you use, how many input tokens you send, how many output tokens you generate, and whether prompt caching applies.

GPT-5.6 Sol Pricing

GPT-5.6 Sol costs $5.00 per 1 million input tokens and $30.00 per 1 million output tokens. It is the most expensive model in the GPT-5.6 family because OpenAI positions Sol as the flagship and most capable model.

Sol should be used when the value of a better answer is higher than the added cost. This includes complex software engineering, architecture planning, advanced reasoning, high-value research, professional analysis, security-sensitive review, and final decision support. For creators, Sol is best used for strategic work: campaign planning, complex story structure, prompt system design, character consistency review, and final production audits.

The key is not to use Sol for everything. If a task only requires a quick title variation or a simple rewrite, Sol may be unnecessary. If a task requires deep reasoning or high accuracy, Sol may be worth the cost.

GPT-5.6 Terra Pricing

GPT-5.6 Terra costs $2.50 per 1 million input tokens and $15.00 per 1 million output tokens. That is half the listed price of Sol for both input and output. OpenAI describes Terra as a strong lower-cost option.

Terra is likely the practical middle model for many applications. It can be used for structured professional workflows where quality matters but the task does not always require the flagship model. For developers, Terra may fit coding support, documentation, summarization, workflow automation, and internal tool generation. For creators, Terra may fit AI video prompt writing, storyboard planning, product video scripts, content briefs, and campaign variations.

Terra is useful when you need strong results at a more manageable cost. Many teams may use Terra as the default model and reserve Sol for harder tasks.

GPT-5.6 Luna Pricing

GPT-5.6 Luna costs $1.00 per 1 million input tokens and $6.00 per 1 million output tokens. OpenAI describes Luna as the fastest and most cost-efficient model in the GPT-5.6 family.

Luna is best for high-volume, lower-risk, latency-sensitive work. This may include classification, routing, short rewriting, simple summaries, metadata generation, prompt variations, caption ideas, title generation, and lightweight content transformations.

For creator tools, Luna can be useful for generating many fast options: 50 hook ideas, 20 video titles, caption variations, short product ad angles, or simple prompt rewrites. For applications, Luna may be used as a first-pass model before escalating harder tasks to Terra or Sol.

The main advantage is cost efficiency. Luna makes sense when speed and scale matter more than maximum capability.

Why Output Tokens Cost More Than Input Tokens

Across all three GPT-5.6 models, output tokens cost more than input tokens. This is common in API pricing because generating output requires model computation during decoding. For cost planning, this means long responses are more expensive than long prompts alone.

For example, a workflow that sends a large prompt but asks for a short classification result may cost less than a workflow that asks for a long report. A creator tool that generates 50 full video scripts will cost more than one that generates 50 short hooks.

This is why product teams should control output length. Use clear formatting, maximum length instructions, and staged workflows. Instead of asking for a huge all-in-one answer, ask for a brief outline first, then expand only the selected version.

Prompt Caching in GPT-5.6

Prompt caching is one of the most important pricing features for GPT-5.6. OpenAI says GPT-5.6 introduces more predictable prompt caching, including explicit cache breakpoints and a 30-minute minimum cache life. For GPT-5.6 and later models, cache writes are billed at 1.25 times the model’s uncached input rate, while cache reads continue to receive a 90% cached-input discount.

In simple terms, prompt caching can reduce costs when you repeatedly send the same or mostly same input prefix. For example, an application may have a long system prompt, brand guide, character bible, product catalog, policy document, or workflow instruction that is reused across many requests. If that stable context can be cached, later requests may be cheaper.

Explicit cache breakpoints are important because they give developers more predictable control over what should be cached. The 30-minute minimum cache life also helps teams design workflows around repeated sessions or batch operations.

How Prompt Caching Helps Creator Workflows

AI video creators often reuse long context. A production system may include a brand style guide, character consistency block, video prompt rules, negative prompt library, product accuracy rules, and shot-list template. Without caching, that repeated context adds cost every time it is sent.

With prompt caching, a creator platform could cache the stable production context and then send only changing details for each shot. For example, the cached portion might include:

brand voice

visual style guide

character bible

prompt rules

negative constraints

shot formatting instructions

quality checklist

Then each request only adds the specific shot: “Shot 4: the character opens the glowing package under blue light.”

This is exactly the kind of workflow where prompt caching can matter. It is not only a developer feature. It can improve the economics of creative production systems.

Cost Strategy: Use the Right Model for the Right Task

The most important GPT-5.6 pricing strategy is model routing. Do not use the most expensive model for every task.

Use Luna for fast, high-volume, low-risk tasks: title ideas, captions, tags, short hooks, simple rewrites, routing, and quick summaries.

Use Terra for balanced production tasks: structured outlines, scripts, prompts, storyboards, documentation, regular coding support, and internal workflow generation.

Use Sol for difficult or high-value tasks: final strategy, complex reasoning, technical architecture, advanced debugging, research-heavy analysis, cybersecurity-sensitive review, and complex creative direction.

This routing strategy can reduce cost without abandoning quality. It lets teams spend more where quality matters and less where speed matters.

Example Pricing Mindset for AI Video Teams

An AI video team might use GPT-5.6 like this:

Luna generates 30 rough video hooks.

Terra expands the best 5 hooks into scripts and shot lists.

Terra creates first-draft AI video prompts.

Sol reviews the final prompt system for consistency and production risks.

Luna generates captions, titles, and metadata.

This workflow uses each model where it fits. Sol is reserved for high-value review. Terra does structured production. Luna handles scale.

A team that used Sol for every hook, caption, and metadata draft would likely spend more than necessary. A team that used Luna for every complex planning task might save money but lose quality. The best approach is balanced.

Cost Control Tips

First, keep stable context cache-friendly. If your application repeatedly uses the same instructions, organize them so prompt caching can help.

Second, control output length. Ask for exactly what you need: 10 hooks, 5 bullet points, a 300-word brief, or a 6-shot list. Vague requests often produce longer, more expensive outputs.

Third, use staged generation. Generate a short outline first, then expand only the chosen option.

Fourth, route by task difficulty. Use Luna for simple scale, Terra for production, and Sol for the hardest work.

Fifth, avoid unnecessary repetition. If a user has already provided a product brief or character bible, store and reuse it efficiently rather than resending unstructured context in every request.

Final Thoughts

GPT-5.6 pricing is built around three model roles. Sol is the most capable and most expensive option at $5.00 input and $30.00 output per 1 million tokens. Terra is the balanced lower-cost option at $2.50 input and $15.00 output. Luna is the fastest and most cost-efficient option at $1.00 input and $6.00 output.

Prompt caching is also central to GPT-5.6 cost planning. OpenAI says GPT-5.6 supports explicit cache breakpoints, a 30-minute minimum cache life, cache writes billed at 1.25 times the uncached input rate, and cached-input reads with a 90% discount.

For teams, the best pricing strategy is simple: use Luna for speed and scale, Terra for balanced production, Sol for the hardest work, and prompt caching for repeated context.

That is how to use GPT-5.6 efficiently without treating every task as a flagship-model task.