GPT-5.6 Sol, Terra and Luna: Availability, Pricing and Cutoff
June 26, 2026
Update, October 18, 2026. I wrote this post on June 26, when GPT-5.6 was a partner-only preview. OpenAI made the family generally available on July 9, 2026, cut prices twice since then, and has released newer models: GPT-6 Astra on September 3, then GPT-6 Sol, GPT-6 Luna and GPT-6.1 Sol in late September. I have updated the availability, pricing and EU sections below to match what OpenAI publishes today. Sources: OpenAI API changelog and ChatGPT What's new.
OpenAI announced the GPT-5.6 family on June 26, 2026. Three models at once: Sol (flagship), Terra (everyday), Luna (fast and cheap). It is the first time they have shipped a three-tier family under a single generation number, and the tier names are meant to stick across future releases.
When I first wrote this, you could not use it yet. That has changed. GPT-5.6 has been generally available in the API since July 9, 2026, and it is in ChatGPT and Codex too. So this is now a practical guide: what each tier knows, what it costs, and where it fits in your work.
Knowledge cutoff and specs
| Spec | Sol | Terra | Luna |
|---|---|---|---|
| Knowledge cutoff | Feb 16, 2026 | Feb 16, 2026 | Feb 16, 2026 |
| Context window | 1,050,000 tokens | 1,050,000 tokens | 1,050,000 tokens |
| Max output | 128,000 tokens | 128,000 tokens | 128,000 tokens |
| API price per 1M tokens (input / output) | $4 / $20 (promotional) | $2 / $12 | $0.20 / $1.20 |
In plain terms: all three tiers know the world up to February 16, 2026. For anything after that, the model needs you to give it the information or it has to search the web. Every tier can read about a million tokens at once and write up to 128,000 tokens in one answer. Very long prompts cost more. Above 272,000 input tokens, OpenAI bills the whole request at 2x input and 1.5x output. The figures come from OpenAI's model pages for Sol, Terra and Luna.
What is GPT-5.6, in one paragraph
GPT-5.6 is a three-tier model family from OpenAI. Sol is the flagship: large context window, strong agentic performance, and competitive per-token pricing. Terra is the middle tier, designed for everyday professional work at GPT-5.5 quality but half the cost. Luna is the fast, affordable option for high-volume routine tasks. The OpenAI help docs describe the tier names as durable capability labels, meaning they plan to carry Sol, Terra, and Luna forward as categories rather than version numbers.
What it is best at
Agentic and multi-step work. Sol scores 88.8% on TerminalBench 2.1, compared to 84.3% for the leading alternative. A new "ultra" subagent mode hits 91.9%. In practice, that means tasks like "research this topic and compile a structured report" or "analyze this dataset and flag the anomalies" complete in one pass. The kind of work that used to need you steering every step now runs to the end on its own. OpenAI's preview page published the benchmarks.
Large documents. Sol's context window is 1,050,000 tokens. That is several long novels. For a due-diligence file plus a regulatory framework plus a client history loaded at once, Sol can hold everything in scope without losing the thread.
Coding and automation for non-developers. If you have been building simple workflows in ChatGPT, writing formulas, or debugging a Zapier step, Sol writes, tests, and debugs in a single pass. OpenAI's preview page lists the coding gains.
Science-adjacent professional work. Sol beats GPT-5.5 on GeneBench v1 while using fewer tokens. If you work in health, pharma, insurance, or research, that efficiency difference adds up when running batches of analysis. OpenAI published the benchmark.
What it is not better at
Writing quality. GPT-5.5 lagged the alternatives in blind prose tests, and we have no GPT-5.6 writing data yet. OpenAI has not published prose quality benchmarks for this release. If natural-sounding writing is central to your use case, treat this as unproven until you can test it yourself.
Finance and legal document reasoning. GPT-5.6 has no public result yet on Hebbia's Finance Benchmark, which tests senior-analyst-level document reasoning. If deep financial or legal document work is your main task, the model has not yet demonstrated it clears the bar. The options with a track record there remain Claude Opus 4.8 and similar alternatives.
Being OpenAI's newest model. In June the catch was access. Today the catch is that newer models sit above it. OpenAI released GPT-6 Astra on September 3 and GPT-6.1 Sol on September 29, 2026, per its API changelog. GPT-5.6 is still a good, cheap workhorse. It is no longer the top of the range.
The three-model menu
| Model | Speed | Depth | Best for | API cost |
|---|---|---|---|---|
| Sol | Slower | Deepest | Complex analysis, autonomous research, large document sets | $4 / $20 |
| Terra | Medium | GPT-5.5 level | Everyday professional work, drafts, data questions | $2 / $12 |
| Luna | Fast | Lighter | Quick questions, summaries, routine intake tasks | $0.20 / $1.20 |
A note for non-developers: on a ChatGPT plan you will not pick between these tiers yourself. OpenAI routes you based on your plan level, the same way Claude routes between Opus and Haiku. If you are building automations, pricing is the real signal. Terra matching GPT-5.5 quality at half the cost is the economics story here. Luna changes the calculation for high-volume small calls.
The privacy catch Dutch professionals should know
Account-level monitoring. During the preview, OpenAI's terms stated that flagged activity triggers "account-level review across relevant conversations and risk signals." That is broader than a single-conversation review. The GDPR data-minimization implication is the same data-minimization question every recent model launch raises: what goes in may be reviewed in more scope than expected. OpenAI's preview page states this.
EU data residency is now possible, but not automatic. In June there was no EU route for GPT-5.6. Now OpenAI lists Sol, Terra and Luna for EU regional processing through eu.api.openai.com, on the Responses, Chat Completions and Batch endpoints. You set residency per project. Europe requires a retention control such as Zero Data Retention or Modified Abuse Monitoring, and those need OpenAI's approval. The residency endpoints also cost 10% more. OpenAI's data controls guide has the details. Microsoft also offers GPT-5.6 in Foundry, including Data Zones Standard deployments, per its launch post.
The structural point is worth saying out loud. Every major model release carries the same catch. A standing data policy is more useful than checking each new release. If you handle client data, employee data, or anything under GDPR scope, get the policy in place once and apply it consistently.
What it costs
Prices have dropped since launch. At launch, Sol listed at $5/$30, Terra at $2.50/$15 and Luna at $1/$6 per 1M tokens input/output. On July 30, OpenAI cut Terra by 20% and Luna by 80%. On August 21, it cut Sol to $4/$20 as promotional pricing that runs at least through November 21, 2026. Today the API prices per 1M tokens are:
- Sol: $4 input, $20 output (promotional)
- Terra: $2 input, $12 output
- Luna: $0.20 input, $1.20 output
Cached input costs a tenth of the normal input price on all three tiers. Luna at these prices changes the math for teams running high-volume classification or extraction tasks. The dates come from OpenAI's API changelog.
On access: GPT-5.6 has been generally available in the API since July 9, 2026. In ChatGPT, Plus and Pro users get GPT-5.6 Sol with a slider that sets how much it reasons. Since early August, Luna is the default model on the Free and Go plans. OpenAI has not published the same plan-by-plan detail for Terra, Business or Enterprise, so check the model picker in your workspace. Source: ChatGPT What's new.
If speed matters more than cost, OpenAI also offers a Fast mode. For Sol it runs up to 2.5x faster at twice the price.
Should you do anything today?
GPT-5.6 is no longer OpenAI's newest model, but it is cheap, stable and widely available. Three concrete steps:
- Test your near-miss tasks. Take tasks you ran on GPT-5.5 or Claude Opus 4.8 that felt "close but not quite" and run them on Sol. Agentic tasks that stalled halfway, large documents that got cut off, automation steps that needed manual correction.
- Route by tier. High-volume, small-scope calls such as summarizing emails, classifying intake forms or extracting structured data belong on Luna. At $0.20 per 1M input tokens, a bigger model is hard to justify for that work. Keep Sol for the hard, multi-step tasks. Decide the routing before you build the workflow. Changing it later is more work.
- Set up EU residency before you send client data. EU processing exists for GPT-5.6 now, but you have to switch it on per project. If you run anything under GDPR scope, set that up and confirm your legal basis and data processing agreement first. Then check whether a GPT-6 model suits your hardest tasks better. OpenAI lists those for EU processing too.
Frequently Asked Questions
When can I use GPT-5.6 in ChatGPT?
Now. GPT-5.6 has been generally available in the API since July 9, 2026, and it is in ChatGPT and Codex. Plus and Pro users get GPT-5.6 Sol, and Luna has been the default model on the Free and Go plans since early August 2026.
What is GPT-5.6's knowledge cutoff?
February 16, 2026, for all three tiers: Sol, Terra and Luna. The model does not know about events after that date unless you give it the information or it searches the web.
Which ChatGPT plans get which tier?
OpenAI says Plus and Pro users get GPT-5.6 Sol, with a slider to control how much it reasons. GPT-5.6 Luna is the default model on the Free and Go plans. OpenAI has not published the same plan-by-plan detail for Terra, Business or Enterprise, so check the model picker in your workspace.
Is Sol better than the alternatives?
It depends on the task. At launch, Sol beat the previous benchmark leader on agentic work (88.8% vs 84.3% on TerminalBench 2.1). Writing quality and finance document reasoning are still unproven in public data. OpenAI has also released newer models since (GPT-6 Astra and GPT-6.1 Sol), so Sol is no longer its top model. For multi-step research and large document sets, Sol is a solid and cheaper choice. For prose-heavy or finance-heavy work, test on your own tasks first.
What is the difference between Sol, Terra, and Luna?
Sol is the flagship tier: strongest reasoning, highest cost. Terra is the everyday tier, positioned at GPT-5.5 quality at a lower price. Luna is the fast, cheap tier for high-volume routine tasks like summarizing, classifying, or extracting data from short inputs. All three share the same 1,050,000-token context window and 128,000-token max output. The tier names are designed to be durable labels that carry forward across future releases.
Is GPT-5.6 safe to use for client data in the EU?
It can be, with the right setup. OpenAI now offers EU regional processing for GPT-5.6 Sol, Terra and Luna through eu.api.openai.com. You set it per project, it requires a retention control such as Zero Data Retention or Modified Abuse Monitoring, and it costs 10% more. Microsoft Foundry also offers GPT-5.6 with Data Zones Standard deployments. Either way, confirm your data processing agreement and legal basis before you send GDPR-regulated data.
Should I switch to GPT-5.6 for my team?
Not wholesale. The pricing is strong, especially Luna for high-volume tasks and Sol for agentic work. A better approach: pick two or three task types where your current model felt limiting, test them on Sol, and route by task rather than switching everything at once. For your hardest work, also compare GPT-6.1 Sol, which OpenAI released on September 29, 2026.
Related articles
Where would AI help your team?
Tell us about your team and the work you want to change. We'll help you pick the right next step.