Skip to main content
  • Pricing
News

GPT-5.6 Release Date, Pricing and Availability: Sol, Terra and Luna Explained

By Oakgen TeamPublished Updated 8 min read

Share
GPT-5.6 Release Date, Pricing and Availability: Sol, Terra and Luna Explained

Yes, GPT-5.6 is live. OpenAI opened a limited preview for trusted partners on June 26, 2026 and released the full family on its API on July 9, 2026. It comes as three models: Sol at $4 input and $20 output per million tokens, Terra at $2 and $12, and Luna at $0.20 and $1.20. Since September, OpenAI's newer GPT-6 models have replaced it as the family OpenAI recommends.

Everything here comes from OpenAI's and Anthropic's developer pages and Oakgen's own code, all read on 2026-10-25. We did not benchmark GPT-5.6 ourselves.

Is GPT-5.6 out yet?

Yes, on the OpenAI API. The table shows how it got there, using OpenAI's own pages.

Date (2026)What happenedOpenAI source
June 26Preview system card published. OpenAI starts "with a limited preview for a small group of trusted partners"GPT-5.6 Preview System Card
July 9GPT-5.6 Sol, Terra and Luna released on the Responses, Chat Completions and Batch APIs. Main system card publishedAPI changelog, GPT-5.6 System Card
July 30Luna price cut by 80%, Terra by 20%API changelog
August 13Ultrafast mode announced, a faster API service tier for SolAPI changelog
August 21Sol price cut to $4 input and $20 output, promotional "at least through November 21, 2026"API changelog, pricing page
September 3GPT-6 Astra releasedAPI changelog
September 22 and 29GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol releasedAPI changelog
October 7Changelog recommends the GPT-6 family for production API useAPI changelog

Inside ChatGPT, we could not confirm which plans show GPT-5.6. OpenAI's ChatGPT help pages blocked our page reader, so check the model picker in your own account. The October 7 changelog entry says the chat-latest API snapshot points to the latest model in ChatGPT for Plus, Pro, Business and Enterprise users, without naming the model.

Why did GPT-5.6 start as a limited preview?

OpenAI says the US government asked for it. The preview system card says "As part of our ongoing engagement with the U.S. government, we previewed our plans" and "At their request, we are starting with a limited preview." It said the models would be generally available "in the coming weeks". The changelog records the full release 13 days later.

Anthropic's Claude Fable 5 also had its access limited soon after launch, and our Claude Fable 5 guide has those dates. If you build on a frontier model, keep a fallback model wired in, because access can change at short notice.

What are GPT-5.6 Sol, Terra and Luna?

They are three models of one generation, priced for different jobs. Sol is the flagship, Terra is the middle option and Luna is the cheapest. They share the same context window, output limit, knowledge cutoff and reasoning settings, so the main difference is price and capability.

GPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.5
OpenAI's description"flagship model for complex professional work""balances intelligence and cost""optimized for cost-sensitive workloads""A new class of intelligence for coding and professional work"
Input / cached / output, per 1M tokens$4 / $0.40 / $20$2 / $0.20 / $12$0.20 / $0.02 / $1.20$5 / $0.50 / $30
Context window1,050,0001,050,0001,050,0001,050,000
Max output128,000128,000128,000128,000
Knowledge cutoffFeb 16, 2026Feb 16, 2026Feb 16, 2026Dec 1, 2025
Reasoning effortnone to maxnone to maxnone to maxnone to xhigh
ReleasedJuly 9, 2026July 9, 2026July 9, 2026April 24, 2026

Prices are OpenAI's standard tier for prompts up to 272K input tokens, read on 2026-10-25. All four accept text and image input and return text.

The system card describes Sol as "our new flagship model", Terra as "a capable lower-cost option" and Luna as "our fastest and most cost-efficient model". It also reports that Sol scored 60.5 on length-adjusted HealthBench Professional against 51.8 for GPT-5.5. That is OpenAI's number, not ours.

How much does the GPT-5.6 API cost?

As of 2026-10-25, Sol costs $4 per million input tokens and $20 per million output tokens, Terra $2 and $12, and Luna $0.20 and $1.20. Cached input costs one tenth of the normal input price on all three.

Three rules change the bill:

  • Long prompts cost more. OpenAI's model pages say "Prompts with >272K input tokens are priced at 2x input and 1.5x output for the full request." Sol over 272K is $8 input and $30 output.
  • Sol's price is promotional. OpenAI says it is "available at least through November 21, 2026" and does not say what follows. Budget for a possible rise after that date.
  • Other service tiers. The pricing page lists separate Batch and Flex rates, which are lower, and Fast mode, the renamed Priority processing, which is higher. Data residency and FedRAMP endpoints add 10% for models released on or after March 5, 2026.

Prices have changed twice since launch, so read the pricing page on the day you commit to a budget. A cheap model is not cheap if its answers need rework, so log cost, latency, failure rate and how often people redo the output for each model you route to.

Is GPT-5.6 better than GPT-5.5?

We have not tested them against each other. On paper, GPT-5.6 Sol costs less than GPT-5.5, knows about events up to February 2026 instead of December 2025, and adds a max reasoning effort level. Context and output limits are the same.

The bigger change is the shape of the lineup. GPT-5.5 is one main model. GPT-5.6 gives you three, so the question shifts from "should I switch?" to "which task goes to which model?" GPT-5.5 is still the practical choice when it is already wired into a working product and stable for your use. OpenAI's deprecations page lists no shutdown for GPT-5.5 as of 2026-10-25, and it names the GPT-5.6 models as replacements for several older models it is retiring.

Is GPT-5.6 ultra mode real?

We could not confirm it, so treat it as unverified. None of the OpenAI pages we read on 2026-10-25 use the term: the three GPT-5.6 model pages, both system cards, the pricing page and the API changelog. Earlier coverage, including ours, described ultra mode as a Sol feature that uses subagents. We have removed that claim.

Two documented features are close to what people mean by it:

  • Max reasoning effort. All three GPT-5.6 models accept reasoning.effort values of none, low, medium (default), high, xhigh and max.
  • Ultrafast mode. Announced August 13, 2026, it is "a new API service tier for GPT-5.6 Sol that runs up to 14x faster than Standard processing." It changes speed, not the model.

The only mention of subagents we found is in the preview system card, where an outside evaluator noted the model coordinating subagents during a test task.

Which GPT-5.6 model should you use for coding agents?

Start with Sol for hard, multi-step agent work, try Terra for everyday coding help, and send Luna the small helper jobs such as summaries, changelog drafts, test names and comment cleanup. Those roles follow OpenAI's own descriptions. Your codebase decides whether Terra is good enough, so test before routing everything to the flagship.

When you test any new coding model, skip toy prompts and use the tasks that break agents in real work:

  1. Fix a bug from an unclear stack trace.
  2. Write the failing test before editing any code.
  3. Move a small feature across three files.
  4. Explain a flaky test and propose how to instrument it.
  5. Review a pull request for hidden behavior changes.
  6. Revise the plan after reading real terminal output.

A model that fails these will not help your workflow, whatever its benchmark scores say. For security work, OpenAI also lists GPT-5.6 Cyber at $12.50 input and $75 output per million tokens, for authorized vulnerability research under its Daybreak program.

What changed when GPT-6 arrived?

GPT-6 is now OpenAI's lead family. OpenAI's models page lists GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna as its flagship models and suggests starting with Astra. GPT-5.6 Cyber is the only GPT-5.6 model left on that page, though the other three still have model pages and prices.

ModelReleased (2026)Input / output per 1M tokensOpenAI's description
GPT-6 AstraSeptember 3$10 / $50"Our most capable model for the most demanding work"
GPT-6.1 SolSeptember 29$2 / $10"Near-Astra performance for complex work at a lower cost"
GPT-6 SolSeptember 22$2 / $10Not on the models page
GPT-6 LunaSeptember 22, generally available October 22$0.10 / $0.50"Our most efficient model for focused, high-volume tasks"
GPT-5.6 SolJuly 9$4 / $20"flagship model for complex professional work"

Prices from OpenAI's pricing page, read on 2026-10-25. On list price alone, GPT-6.1 Sol costs half of GPT-5.6 Sol and GPT-6 Luna costs half of GPT-5.6 Luna. If you are choosing a model for new work today, evaluate GPT-6 first. Stay on GPT-5.6 if it already passes your tests and you do not want to re-validate yet.

What are the alternatives to GPT-5.6?

Within OpenAI, the alternatives are the GPT-6 models above. Outside OpenAI, Anthropic's current lineup has similar tiers. Anthropic's models overview, read on 2026-10-25, lists:

JobOpenAI optionAnthropic optionAnthropic price per 1M tokens
Hardest reasoning and long agent runsGPT-6 Astra or GPT-5.6 SolClaude Fable 5.1$10 / $50
Most production workGPT-6.1 Sol or GPT-5.6 TerraClaude Opus 5.5$4 / $20
Fast everyday workGPT-5.6 TerraClaude Sonnet 5.5$2 / $10
High-volume classification and extractionGPT-6 Luna or GPT-5.6 LunaClaude Haiku 5.5from $0.10 / $0.50

All four Claude models have a 1M token context window and 128K max output. Anthropic says to start with Opus 5.5 "for most workloads". For how Fable compares with GPT-5.6 Sol on price and specs, read our Claude Fable 5 guide.

Most writing, analysis, code review, planning and research work does not need to wait for any one model. Waiting makes sense when you are choosing a model for a production routing decision and need your own test results, since you cannot judge a model from someone else's screenshot.

Can you use GPT-5.6 on Oakgen?

No. Oakgen Agent Chat does not offer GPT-5.6 or any other OpenAI chat model. As of 2026-10-25 it offers Oakgen's own tiers: Oakgen Auto (the default), Oakgen Max and Oakgen Pro Max. Per token, Max costs twice the rate of Auto and Pro Max five times.

Agent Chat needs an active Oakgen subscription and bills chat in credits per token. If you like the Sol, Terra and Luna way of working, the same split works here: Pro Max for high-stakes reasoning, Max for most work and Auto for quick drafts. That is an analogy for routing, not GPT-5.6 on Oakgen. To decide which lane a task belongs in, ask:

  • Is it high-stakes, and would a wrong answer cause rework?
  • Does it need tools or several steps of reasoning?
  • Does speed matter more than polish?
  • Is it one task or thousands?

When you compare answers between models, judge accuracy, whether the reasoning holds up, whether it says when it is unsure, formatting, cost and speed, and how much editing it needs.

What Agent Chat adds over a plain chat window is creation. The same conversation can call Oakgen's image, video and audio models, so a plan written in chat can become a product image, a short video or a voiceover.

Plan and create in Oakgen Agent Chat

Use Oakgen Auto, Max or Pro Max for the thinking, then generate images, video and audio in the same chat.

Open Agent Chat

Which OpenAI models can you use on Oakgen?

OpenAI's image models. The image generator has GPT Image 2 and GPT Image 2.5 in two styles, Flare and Sunburst, each with an edit version. Credit prices as of 2026-10-25:

ModelLow qualityMedium qualityHigh quality
GPT Image 2, 1024x10242 credits18 credits72 credits
GPT Image 2, 1024x1536 or 1536x10242 credits14 credits56 credits
GPT Image 2.5 Flare or Sunburst, 1024x10245 credits7 credits21 credits

The generator shows the credit price before you run. For how the two generations differ, see GPT Image 2.5 vs GPT Image 2. Plans and credit add-ons are on the pricing page, and plans can change.

When is another tool a better choice?

  • If you want GPT-5.6 or GPT-6 itself, use OpenAI's API or ChatGPT. Oakgen does not resell OpenAI chat models.
  • If you need a model inside your own product, call the vendor's API directly. Oakgen Agent Chat is a workspace, not an API.
  • If you only chat and never generate images, video or audio, a single-vendor chat subscription is the more direct route.
  • If you need the cheapest tokens for bulk extraction, GPT-6 Luna at $0.10 and $0.50 per million tokens is cheaper than anything in Agent Chat.

How we checked this

On 2026-10-25 we read OpenAI's GPT-5.6 preview system card and main system card, the API changelog, the pricing page, the models page, the deprecations page and the model pages for GPT-5.6 Sol, Terra, Luna and GPT-5.5. We read Anthropic's models overview for the Claude prices. OpenAI's openai.com and help.openai.com pages refused our page reader, so we could not confirm ChatGPT plan access, and we cut claims that relied on them, including ultra mode and the old claim that Terra matches GPT-5.5 at half the price.

Oakgen facts come from Oakgen's Agent Chat model list and its generated credit price table, read the same day. We did not run GPT-5.6, GPT-5.5 or GPT-6 side by side for this article. Benchmark figures are OpenAI's own. Vendor prices and Oakgen credit prices change, so check the vendor page and Oakgen's pricing page before you commit.

Frequently asked questions

When was GPT-5.6 released?

OpenAI published the GPT-5.6 preview system card on June 26, 2026, which says GPT-5.6 started as a limited preview for a small group of trusted partners. OpenAI's API changelog records the release of the full GPT-5.6 family, Sol, Terra and Luna, on July 9, 2026, the same day the main system card was published. All three are on the OpenAI API as of 2026-10-25.

Is GPT-5.6 still OpenAI's newest model?

No. OpenAI released GPT-6 Astra on September 3, 2026, GPT-6 Sol and GPT-6 Luna on September 22, and GPT-6.1 Sol on September 29, per its API changelog. On October 7 the changelog started recommending the GPT-6 family for production API use. GPT-5.6 Sol, Terra and Luna are still listed on OpenAI's pricing page as of 2026-10-25, and the deprecations page shows no shutdown date for them.

How much does GPT-5.6 cost?

Per million tokens on OpenAI's standard tier, as of 2026-10-25: GPT-5.6 Sol costs $4 input and $20 output, Terra $2 and $12, and Luna $0.20 and $1.20. Cached input is one tenth of the input price. Prompts over 272K input tokens are billed at 2x input and 1.5x output for the whole request. OpenAI says Sol's current price is promotional and runs at least through November 21, 2026.

What is GPT-5.6 ultra mode?

We could not confirm that a feature called ultra mode exists. None of the OpenAI pages we read on 2026-10-25 mention it: the GPT-5.6 model pages, the two system cards, the pricing page and the API changelog. What OpenAI does document is a max reasoning effort setting on all three GPT-5.6 models, and Ultrafast mode, an API service tier for GPT-5.6 Sol announced August 13, 2026 that OpenAI says runs up to 14x faster than Standard processing.

Can I use GPT-5.6 on Oakgen?

No. As of 2026-10-25 Oakgen Agent Chat offers Oakgen's own chat tiers, Oakgen Auto, Oakgen Max and Oakgen Pro Max, not OpenAI chat models. It needs an active subscription. Oakgen does offer OpenAI's image models GPT Image 2 and GPT Image 2.5 in its image generator, priced in credits.

Sources

  1. OpenAI, GPT-5.6 Preview System Card (June 26, 2026)
  2. OpenAI, GPT-5.6 System Card (July 9, 2026)
  3. OpenAI API docs, changelog
  4. OpenAI API docs, pricing
  5. OpenAI API docs, models
  6. OpenAI API docs, GPT-5.6 Sol model page
  7. OpenAI API docs, GPT-5.6 Terra model page
  8. OpenAI API docs, GPT-5.6 Luna model page
  9. OpenAI API docs, GPT-5.5 model page
  10. OpenAI API docs, deprecations
  11. Claude Platform docs, models overview
Share