OpenAI API pricing now revolves around the GPT-5.6 family—Sol, Terra, and Luna text tiers plus Cyber, a cybersecurity specialist—spanning input/output rates from $0.20/$1.20 up to $12.50/$75, a 60x gap (source). This post breaks down the four models' unit prices, cache discounts, top-up flow, rate limits, and integration setup so you can decide whether it's worth plugging in.
GPT-5.6 Family Pricing: Four Tiers Spanning 60x
From highest to lowest input/output rates the GPT-5.6 family runs Cyber, Sol, Terra, and Luna—all at 1,050,000 context (Cyber is short-context only) with up to 128K output.
| Model | Model ID | Input | Output | Cache-Hit Input | Context |
|---|---|---|---|---|---|
| GPT-5.6 Cyber | gpt-5.6-cyber | $12.50 / M | $75 / M | $1.25 / M | ≤270K |
| GPT-5.6 Sol | gpt-5.6-sol | $5 / M | $30 / M | $0.50 / M | 1,050,000 |
| GPT-5.6 Terra | gpt-5.6-terra | $2 / M | $12 / M | $0.20 / M | 1,050,000 |
| GPT-5.6 Luna | gpt-5.6-luna | $0.20 / M | $1.20 / M | $0.02 / M | 1,050,000 |
These are Standard-processing list prices for contexts under 270K (source). Cyber is the most expensive but only serves security use cases—day-to-day decisions happen across the other three tiers.
GPT-5.6 Sol: Current API Flagship
Sol is the default flagship (model ID gpt-5.6-sol, aliased as gpt-5.6) for complex reasoning, professional coding, and multi-step agents. It supports reasoning.effort from none to xhigh plus the full toolchain—Functions, Web search, File search, Computer use, Code interpreter—through both Chat Completions and the Responses API. At $5/$30 it's the strongest and the priciest.
GPT-5.6 Terra: The Balanced Workhorse
Terra sits between Sol and Luna at $2/$12 with the same full toolchain and an uncompromised 1,050,000 context. When Sol's unit cost is too high but Luna isn't enough, Terra is usually the production API sweet spot.
GPT-5.6 Luna: Fast and Cheap
Luna costs just $0.20 input and $1.20 output per million tokens, with cache-hit input as low as $0.02—the cheapest of the four. It fits high-frequency light completions, routing and classification, batch formatting, and sub-agents. "Start on Luna, escalate hard tasks to Sol" is the classic cost-saving architecture on official pricing.
GPT-5.6 Cyber: Security Specialist
Cyber is the Daybreak-series cybersecurity model at $12.50/$75, supporting short context only (≤270K). It targets vulnerability analysis, threat detection, and security agents—for general coding and reasoning, go straight to Sol instead.
Caching, Batch, and Long Context: Where the Real Savings Live
Cache-hit input runs at a tenth of list price, Batch API knocks off roughly 50%, and prompts over 272K tokens bill the whole request at a multiplier—these levers decide your actual invoice.
- Cache-hit input: Sol $0.50/M, Terra $0.20/M, Luna $0.02/M—turn on prompt caching for repeated system prompts, long conversation history, and RAG prefixes (source)
- Cache writes billed separately: Sol $6.25/M, Terra $2.50/M, Luna $0.25/M, Cyber $15.625/M
- Batch API: roughly 50% off input and output for delay-tolerant offline jobs
- Long context: prompts over 272K input tokens bill the full request at 2x input and 1.5x output
- Regional markup: Data residency adds 10% for the GPT-5.6 family
- Web Search tool: $10 per 1K calls, with search-content tokens free
Top-Up, Official Entry, and API Key Setup
The OpenAI API official entry is platform.openai.com—after signup you create a project, add billing, and generate an API key; Playground and production calls share one balance, and ChatGPT subscription quota doesn't count.
- Top-up: pay-as-you-go, top up first and you're billed from balance; service stops when balance hits zero, so top up what you need
- API key: platform.openai.com → create a project → generate a key, then configure it in Codex CLI, Cursor, Cline, or your own backend's environment variables
- The boundary: ChatGPT web subscriptions (see ChatGPT) live on chatgpt.com—their quota never offsets API bills, two fully separate lines
- Quotas, usage, and rate limits: balance is the hard cap; different models and endpoints have their own default rate limits, with project usage limits and upgrade paths in the console
Is It Worth It: Match the Model to Your Workload
Whether GPT-5.6 is worth it depends on your call volume and cost sensitivity—route high-frequency light work through Luna, keep Terra as the production mainstay, spend on Sol only for complex reasoning and coding, and leave Cyber to security tasks.
| Workload | Recommended Model | Input | Output | Why |
|---|---|---|---|---|
| Sub-agents, bulk completion | GPT-5.6 Luna | $0.20 / M | $1.20 / M | 25x cheaper than Sol, good enough |
| Production mainstay, mid complexity | GPT-5.6 Terra | $2 / M | $12 / M | Balanced capability and cost |
| Complex reasoning, pro coding | GPT-5.6 Sol | $5 / M | $30 / M | Strongest API capability today |
| Security analysis, threat detection | GPT-5.6 Cyber | $12.50 / M | $75 / M | Domain-specialized |
Versus a subscription: a subscription trades a fixed monthly fee for capped quota (Plus at $20/month), while the API bills per token with no cap. For light daily chat and casual agents, the subscription is more convenient; for high-volume calls where you want to control the cost ceiling, the API is more transparent.
Cross-vendor, Claude API flagship sits at $10/$50, DeepSeek flagship around $0.435/$0.87, and you can switch many models under one key via OpenRouter. GPT-5.6 prices high within the international first tier, but the Luna tier holds its own against low-cost models.
Bottom Line
- The official pricing page https://platform.openai.com/docs/pricing is the single source of truth—watch cache-hit rates and Batch discounts for ongoing use.
- Estimate monthly call volume before picking a model; don't route everything through Sol.
- Contexts over 272K bill the whole request at 2x/1.5x, so design your input structure up front.
- API and ChatGPT subscriptions are separate books—don't treat plan quota as API balance.