{"id":2728,"date":"2026-07-31T06:09:45","date_gmt":"2026-07-31T06:09:45","guid":{"rendered":"https:\/\/dmarketertayeeb.com\/blog\/openai-gpt-5-6-luna-terra-price-cuts\/"},"modified":"2026-09-05T14:35:37","modified_gmt":"2026-09-05T14:35:37","slug":"openai-gpt-5-6-luna-terra-price-cuts","status":"publish","type":"post","link":"https:\/\/dmarketertayeeb.com\/blog\/openai-gpt-5-6-luna-terra-price-cuts\/","title":{"rendered":"GPT-5.6 Pricing 2026: Current Sol, Terra and Luna API and Codex Rates"},"content":{"rendered":"

Short answer:<\/strong> OpenAI\u2019s current GPT-5.6 rates are $4 input \/ $0.40 cached input \/ $20 output<\/strong> per million tokens for GPT-5.6 Sol, $2 \/ $0.20 \/ $12<\/strong> for GPT-5.6 Terra, and $0.20 \/ $0.02 \/ $1.20<\/strong> for GPT-5.6 Luna on the documented Work\/Codex token-based rate card. OpenAI says the Sol price is promotional and available at least through November 21, 2026<\/strong>. The July 30 Luna and Terra reductions remain part of the current pricing story; the August 21 update is the important correction for Sol.<\/p>\n\n\n\n

These are not a promise that a ChatGPT subscription is billed like a simple API invoice. Included plan usage, 5-hour and weekly limits, legacy credit meters, enterprise billing and API-key traffic can follow different rules. The useful budgeting unit is cost per accepted result after model usage, tools, retries and human review\u2014not price per message.<\/p>\n\n\n\n

Current rates and dated changes are linked to the OpenAI model catalog<\/a>, rate card<\/a> and GPT-5.6 announcements, accessed September 5, 2026.<\/p>\n\n\n\n\n

The current GPT-5.6 rate table<\/h2>\n\n\n\n
\n\n
Model<\/th>Input \/ 1M tokens<\/th>Cached input \/ 1M<\/th>Output \/ 1M<\/th>OpenAI\u2019s stated role<\/th><\/tr><\/thead>\n
GPT-5.6 Sol<\/strong><\/td>$4.00<\/td>$0.40<\/td>$20.00<\/td>Flagship model for complex professional work<\/td><\/tr>\n
GPT-5.6 Terra<\/strong><\/td>$2.00<\/td>$0.20<\/td>$12.00<\/td>Balance of intelligence and cost<\/td><\/tr>\n
GPT-5.6 Luna<\/strong><\/td>$0.20<\/td>$0.02<\/td>$1.20<\/td>Cost-sensitive, high-volume workloads<\/td><\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n

The current ChatGPT rate card<\/a> lists these rates for supported ChatGPT Work and Codex activity and notes that the Sol promotional pricing is available at least through November 21. The live OpenAI model catalog<\/a> and model pages provide the corresponding API model details. Check the billing surface and account or workspace policy before using the table in a purchase decision.<\/p>\n\n\n\n\n

Should you also compare GPT-6 Astra?<\/h2>\n\n\n\n

GPT-6 Astra has a different rate structure from GPT-5.6 Sol, Terra and Luna, so budget it as a separate option. OpenAI\u2019s current Astra model page<\/a> lists the following standard API rates per one million tokens:<\/p>\n\n\n\n

\n\n
Model<\/th>Input<\/th>Cached input<\/th>Cache write<\/th>Output<\/th><\/tr><\/thead>\n
GPT-6 Astra<\/strong><\/td>$10.00<\/td>$1.00<\/td>$12.50<\/td>$50.00<\/td><\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n

For Astra API requests, OpenAI says input above 272,000 tokens is charged at 2\u00d7 the input and cached-input rates and 1.5\u00d7 the output rate for the full request. Cache writes are listed at 1.25\u00d7 uncached input. Batch and Flex processing are 50% of standard API rates, while Fast processing is 2\u00d7 applicable API rates where available. The Enterprise Chat, Work and Codex rate card<\/a> is a separate surface: it currently lists Fast at 2.5\u00d7 there and describes Codex Astra without the API page\u2019s >272K multiplier or separate cache-write charge. Do not use it as an interchangeable API multiplier.<\/p>\n\n\n\n

For GPT-5.6 pricing history and budget examples, use this page. Use the GPT-5.6 model-selection guide<\/a> for prompting and routing. Use the dedicated Astra pricing guide<\/a> for Astra-specific estimates. Treat Astra\u2019s price premium as a reason to test whether its extra capability clears the acceptance bar. The latest-model guidance<\/a> describes Astra\u2019s rollout and API behavior.<\/p>\n\n\n

What changed, and when?<\/h2>\n\n\n\n

June 26: the preview rates<\/h3>\n\n\n\n

OpenAI\u2019s GPT-5.6 preview page listed the original rates: Sol at $5 input\/$30 output, Terra at $2.50\/$15 and Luna at $1\/$6 per million tokens. Those figures are useful historical context, but they are not the current table.<\/p>\n\n\n\n

July 30: Luna and Terra became cheaper<\/h3>\n\n\n\n

In its July 30 price-performance announcement<\/a>, OpenAI announced an 80% reduction for Luna and a 20% reduction for Terra. The company said the lower Luna and Terra prices also affect how usage is counted in eligible ChatGPT Work and Codex activity. It said subscription prices and quota budgets did not change. The same announcement introduced Fast mode for the API, with up to 2.5 times Standard processing speed at twice the Standard price for Sol and no change in intelligence.<\/p>\n\n\n\n

August 21: Sol received a temporary price reduction<\/h3>\n\n\n\n

OpenAI\u2019s current GPT-5.6 page<\/a> includes an August 21 update saying API and credit pricing for Sol fell by more than 20% for the next three months. The live Sol model page<\/a> now shows $4 input, $0.40 cached input and $20 output, and says the promotional pricing is available at least through November 21. This supersedes the July 30 \u201cSol pricing remains unchanged\u201d statement for current budgeting while preserving that statement as a dated historical record.<\/p>\n\n\n\n

The end-date language matters. \u201cAvailable at least through\u201d does not tell us what the price will be after November 21, so do not build a long-term forecast on an assumed permanent reduction.<\/p>\n\n\n\n

What these prices do\u2014and do not\u2014cover<\/h2>\n\n\n\n

Token rates are easiest to interpret when you identify the product surface first.<\/p>\n\n\n\n

\n\n
Surface<\/th>What the current evidence supports<\/th>Do not assume<\/th><\/tr><\/thead>\n
OpenAI API<\/strong><\/td>Model pages list token rates, context rules, endpoints and organization rate limits.<\/td>That an API request follows a ChatGPT subscription allowance.<\/td><\/tr>\n
ChatGPT Work and Codex<\/strong><\/td>The rate card lists supported token-based model rates and the July announcement says Luna\/Terra use fewer credits in eligible activity.<\/td>That every plan, workspace or legacy meter uses the same accounting.<\/td><\/tr>\n
ChatGPT Chat<\/strong><\/td>The ChatGPT product has its own plan, model-picker and usage experience.<\/td>That a Chat release or API rate automatically changes Work or Codex behavior.<\/td><\/tr>\n
Enterprise and legacy billing<\/strong><\/td>The rate card distinguishes token-based enterprise pricing, included usage and legacy rates.<\/td>That one public table determines every contract or workspace invoice.<\/td><\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n

OpenAI\u2019s current Codex Help Center guidance<\/a> says Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents can share an agentic allowance and credit pool when available on a plan. It also says usage depends on model, surface, task complexity, context, reasoning, speed and tools. A subscription user should check the usage dashboard or limit banner rather than converting a token rate into a fixed number of messages.<\/p>\n\n\n\n

How to calculate an illustrative API job<\/h2>\n\n\n\n

For a simple job below the long-context threshold, use:<\/p>\n\n\n\n

(input tokens \u00d7 input rate) + (cached input tokens \u00d7 cached rate) + (output tokens \u00d7 output rate)<\/code><\/p>\n\n\n\n

For example, 100,000 uncached input tokens and 10,000 output tokens would be approximately:<\/p>\n\n\n\n

\n\n
Model<\/th>Arithmetic<\/th>Illustrative token cost<\/th><\/tr><\/thead>\n
GPT-5.6 Sol<\/td>(0.1 \u00d7 $4) + (0.01 \u00d7 $20)<\/td>$0.60<\/td><\/tr>\n
GPT-5.6 Terra<\/td>(0.1 \u00d7 $2) + (0.01 \u00d7 $12)<\/td>$0.32<\/td><\/tr>\n
GPT-5.6 Luna<\/td>(0.1 \u00d7 $0.20) + (0.01 \u00d7 $1.20)<\/td>$0.032<\/td><\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n

These figures are arithmetic illustrations, not an estimate of the total cost of a production workflow. They exclude tool-specific fees, retries, cached-input savings, long-context multipliers, storage, orchestration and human review. A model that is cheaper per token can be more expensive per accepted result if it fails the quality bar more often.<\/p>\n\n\n\n

Long context, caching and Fast mode<\/h2>\n\n\n\n

Long context<\/h3>\n\n\n\n

The current GPT-5.6 Sol page<\/a> says requests above 272,000 input tokens receive a higher multiplier for the full request: twice the input rate and 1.5 times the output rate. The exact treatment can vary by model and billing page, so check the applicable documentation before sending a very large context. The safest cost control is often a compact source extract, not a larger context window.<\/p>\n\n\n\n

Prompt caching<\/h3>\n\n\n\n

The current rate table exposes a lower cached-input rate. Stable prefixes, project instructions and repeated source context may qualify for caching according to the applicable API behavior. Do not assume a cache hit; measure it in the request logs or billing data available to your account.<\/p>\n\n\n\n

Fast mode<\/h3>\n\n\n\n

OpenAI describes Fast mode as an API processing option for Sol that can deliver up to 2.5 times Standard speed at twice the Standard price, without changing model intelligence. Fast mode is easier to justify when a human is blocked on an urgent task or a delay changes the business outcome. It is usually harder to justify for an overnight batch, background enrichment or a queue with no waiting reviewer. \u201cUp to\u201d is not a guarantee for every workload.<\/p>\n\n\n\n

Route models by cost per accepted result<\/h2>\n\n\n\n

Use the model role as a starting hypothesis, then test a representative sample against a clear acceptance bar. The cost-per-accepted-result guide<\/a> provides a related way to compare model economics after quality and review are counted.<\/p>\n\n\n\n

\n\n
Workflow step<\/th>Starting model<\/th>Why<\/th>Control<\/th><\/tr><\/thead>\n
Query, metadata or field classification<\/td>Luna<\/td>High volume and mechanically testable<\/td>Labelled sample, schema validation and drift review<\/td><\/tr>\n
First-pass brief or routine synthesis<\/td>Terra<\/td>Balances evidence integration and cost<\/td>Source links and editor review<\/td><\/tr>\n
Conflicting analytics diagnosis<\/td>Sol<\/td>Several causal explanations and high decision cost<\/td>Verified exports, assumptions and sign-off<\/td><\/tr>\n
Urgent interactive investigation<\/td>Sol, optionally Fast<\/td>Latency may change the outcome<\/td>Measure elapsed time and incremental value<\/td><\/tr>\n
Routine implementation after the design is settled<\/td>Luna or Terra<\/td>Bounded changes can be checked automatically<\/td>Tests, diff review and rollback path<\/td><\/tr>\n<\/tbody>\n<\/table><\/figure>\n\n\n\n

For model-selection and prompting patterns, see the GPT-5.6 model guide<\/a>. For usage windows, credits and Codex workflow controls, see the Luna Max Codex guide<\/a>.<\/p>\n\n\n\n

A seven-step pricing and routing checklist<\/h2>\n\n\n\n
    \n
  1. Write down the surface.<\/strong> Record API, ChatGPT Chat, Work, Codex, enterprise token billing or legacy credits.<\/li>\n
  2. Record the current model ID.<\/strong> Do not rely on a nickname or a stale dashboard label.<\/li>\n
  3. Measure the input shape.<\/strong> Separate uncached input, cached input, output, tool calls and long-context requests.<\/li>\n
  4. Define acceptance.<\/strong> State the accuracy, format, latency, safety or human-review threshold before routing cheaper.<\/li>\n
  5. Test a representative sample.<\/strong> Compare Luna, Terra and Sol only on the work each is expected to perform.<\/li>\n
  6. Include review and correction cost.<\/strong> Rejected output, retries and a bad decision can dominate token price.<\/li>\n
  7. Recheck date boundaries.<\/strong> Record the current Sol promotional boundary and refresh the rate table when OpenAI changes it.<\/li>\n<\/ol>\n\n\n\n

    What this means for digital marketing teams<\/h2>\n\n\n\n

    The July and August pricing changes make more supporting work economical, but they do not authorize low-quality content at scale. A governed SEO workflow can use Luna for query grouping and link checks, Terra for source extraction and outline options, and Sol for duplicate-intent adjudication or a high-consequence recommendation. Human review still owns claims, publication, spend and customer-facing changes.<\/p>\n\n\n\n

    For paid media, use Luna to label search terms and creative variants, Terra to summarize stable exports and Sol to reconcile attribution, conversion lag and budget constraints. For analytics, start with deterministic cleanup, then escalate when metric definitions or causal explanations conflict. The ChatGPT Work scheduled-tasks guide<\/a> covers permissions, deduplication and read-back controls that remain important regardless of model price; the ChatGPT Work and Codex workflow guide<\/a> covers the adjacent workspace architecture.<\/p>\n\n\n\n

    Frequently asked questions<\/h2>\n\n\n\n

    What is the current GPT-5.6 Sol price?<\/h3>\n\n\n\n

    The current model page and ChatGPT rate card show $4 per million input tokens, $0.40 per million cached input tokens and $20 per million output tokens for the stated promotional period. OpenAI says this pricing is available at least through November 21, 2026. Check the applicable account and billing surface.<\/p>\n\n\n\n

    What are the current GPT-5.6 Terra and Luna prices?<\/h3>\n\n\n\n

    Terra is $2 input, $0.20 cached input and $12 output per million tokens. Luna is $0.20 input, $0.02 cached input and $1.20 output per million tokens on the current documented table.<\/p>\n\n\n\n

    Did GPT-5.6 Sol become cheaper permanently?<\/h3>\n\n\n\n

    OpenAI describes the current Sol rate as promotional and says it is available at least through November 21. The post-promotion rate is not specified, so treat the current figure as date-bound.<\/p>\n\n\n\n

    Did ChatGPT or Codex subscriptions become cheaper?<\/h3>\n\n\n\n

    OpenAI\u2019s July 30 announcement said ChatGPT and Codex subscription prices and quota budgets did not change. Lower model and credit rates can make eligible usage go further, but that is not the same as a lower monthly subscription price or a fixed quota increase.<\/p>\n\n\n\n

    Does the price table tell me how many messages I get?<\/h3>\n\n\n\n

    No. Usage depends on the model, context, reasoning, tools, surface and account rules. The Codex Help Center recommends checking the usage dashboard or limit notice for the allowance and options that apply to you.<\/p>\n\n\n\n

    Is Fast mode a smarter model?<\/h3>\n\n\n\n

    No. OpenAI describes Fast mode as a speed-priced processing option for Sol. It can be useful when latency has measurable value, but it does not change the model\u2019s intelligence.<\/p>\n\n\n\n

    Bottom line<\/h2>\n\n\n\n

    The current GPT-5.6 economics are materially different from the June preview table and from the July 30 Sol line. Use $4\/$0.40\/$20 for Sol, $2\/$0.20\/$12 for Terra and $0.20\/$0.02\/$1.20 for Luna on the documented current rate card, and date the Sol promotion because OpenAI only promises it at least through November 21, 2026. Route each workflow to the least expensive model that passes a representative acceptance test, and keep API, ChatGPT, Work, Codex and legacy billing rules separate.<\/p>\n\n\n\n

    Current GPT-5.6 rates and dated changes are documented on the linked Sol<\/a>, Terra<\/a> and Luna<\/a> pages, accessed September 5, 2026.<\/em><\/p>","protected":false},"excerpt":{"rendered":"

    See current GPT-5.6 Sol, Terra and Luna input, cached-input and output rates, the Sol promotion, Fast mode, Codex boundaries and a bounded GPT-6 Astra comparison.<\/p>\n","protected":false},"author":1,"featured_media":2727,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[183,180,178],"tags":[194,319,320,299,296],"class_list":["post-2728","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-in-marketing","category-ai-news","category-artificial-intelligence","tag-ai-marketing","tag-ai-workflows","tag-codex","tag-gpt-5-6","tag-openai","has-featured-image"],"_links":{"self":[{"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/posts\/2728","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/comments?post=2728"}],"version-history":[{"count":4,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/posts\/2728\/revisions"}],"predecessor-version":[{"id":2926,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/posts\/2728\/revisions\/2926"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/media\/2727"}],"wp:attachment":[{"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/media?parent=2728"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/categories?post=2728"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/dmarketertayeeb.com\/blog\/wp-json\/wp\/v2\/tags?post=2728"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}