What changed in GPT-5.6 Sol
The update, detailed in OpenAI's official blog post and release notes, applies only to the ChatGPT chat experience — it does not touch GPT-5.6 Sol in ChatGPT Work or Codex. The changes fall into three buckets:
More focused answers. The model now trims unnecessary detail and excess formatting. For quick questions, it delivers a direct answer. For complex tasks like multi-step planning or research, it gives a fuller response while keeping the main recommendation visible. OpenAI's own example — a bike-ride weather check — shows Sol cutting a six-paragraph answer from GPT-5.5 Instant down to three sentences plus an optional alert offer.
Better factual accuracy. In an internal evaluation across finance, medicine, and law prompts, responses containing at least one factual error were about 62% less common with GPT-5.6 Luna and 68% less common with GPT-5.6 Sol compared with GPT-5.5 Instant. These are self-reported figures and have not been independently verified, but the direction is consistent with what we have observed running inference workloads in production: each generation brings measurable improvements in factual grounding, particularly on structured knowledge tasks.
A reasoning-depth slider. Plus and Pro users can now choose how much "thought" ChatGPT puts into an answer. The slider has five levels — Instant, Medium, High, Extra High (Pro only), and the Pro-standard tiers — mapped from recommendations an OpenAI staffer published last month. Lower settings suit everyday questions; higher settings are meant for planning, research, writing, and coding. The slider works on web, mobile, and desktop.
Free users get Luna — and a "Think" button that is not what it sounds like
GPT-5.6 Luna becomes the default model for Free and Go users starting this week. Alongside it, free users will get unlimited text chats (subject to abuse guardrails) and a new "Think" button for harder questions — but this button only gives Luna more time to reason, rather than switching to a stronger model like Sol.
OpenAI pushed back on the framing that free users are being "restricted," arguing that Luna is still a step up from GPT-5.5 Instant, the previous free default. In absolute terms, that is true: Luna's 62% factual-error reduction over Instant is a genuine improvement. But the larger point stands — free users no longer get any access to OpenAI's frontier reasoning models. That was not always the case, and the quality gap between Luna and Sol is real.
Limits on file uploads, image generation, and other tools remain in place for free users.
EUR pricing: what each tier actually costs
OpenAI's pricing page lists plans in USD, but European users are charged in euros. Here is the breakdown at the current exchange rate (approximately 1 USD = 0.87 EUR as of August 8, 2026):
| Plan | USD/month | EUR/month (approx.) | Default model | Reasoning slider |
|---|---|---|---|---|
| Free | €0 | €0 | GPT-5.6 Luna | No (Think button only) |
| Go | $10 | ~€8.70 | GPT-5.6 Luna | No (Think button only) |
| Plus | $20 | ~€17.40 | GPT-5.6 Sol | Yes |
| Pro | $200 | ~€174 | GPT-5.6 Sol Pro | Yes (incl. Pro tiers) |
The gap between free/Luna and paid/Sol is substantial. At roughly €17.40 per month, Plus remains the most practical tier for European power users who need reliable factual answers — a price point that sits between Claude Fable 5's ~€22 and Gemini Advanced's ~€22. Go, at ~€8.70, gets you more uploads and tool usage but still uses Luna as the default model. Pro at ~€174 is firmly aimed at developers and researchers who can justify the cost with heavy usage.
What the slider actually means in practice
The reasoning slider is the most interesting part of this update from a technical standpoint. It lets users dial up or down how much compute ChatGPT spends on a response — effectively a user-facing abstraction over test-time compute scaling. Lower settings give faster, cheaper answers; higher settings burn more tokens for deeper reasoning.
In our own production experience running multiple AI pipelines — from article generation to code review — this kind of dynamic reasoning control can make a real difference in cost efficiency. When we need a quick fact check, the lightweight path is fine. When we are analysing a research paper or debugging a pipeline, the heavier reasoning is worth the extra latency. Having both in one interface, with consistent tone, beats the old model-switcher approach most users ignored anyway.
That said, the slider adds complexity. An OpenAI employee recommended users start one level lower than they expect and scale up only when needed. For most everyday use, "Instant" will likely suffice. The real value kicks in when you know you are working on something that requires careful verification — contract review, academic research, or multi-step financial analysis.
European angle: AI Act, data residency, and availability
GPT-5.6 Sol and Luna are available across the EU. ChatGPT has been continuously available in the European Economic Area, and OpenAI offers data residency in the EU for Enterprise customers (not for individual plans like Plus or Pro). For European users on consumer plans, data is processed in OpenAI's US infrastructure, which matters for GDPR compliance — particularly for businesses that might use ChatGPT for work-related tasks.
Under the EU AI Act, which entered its main enforcement phase on August 2, 2026, general-purpose AI models like GPT-5.6 must meet transparency obligations, including providing summaries of training data and implementing measures to comply with copyright law. OpenAI published a compliance statement on July 31 but notably omitted the copyright chapter. For European developers and companies building on top of these models, this regulatory gap is worth tracking.
Pricing in EUR is consistent across EU member states — there is no regional price discrimination within the bloc. That is worth noting because some AI services (looking at you, Claude) have historically charged different prices depending on the country.
Should you upgrade from Free to Plus?
If you use ChatGPT for anything beyond casual conversation — research, work, studying, or coding — the ~€17.40 monthly cost of Plus is almost certainly worth it. The factual accuracy difference between Sol and Luna is real, the reasoning slider gives you control over response quality, and the expanded context window (54K for Plus vs. 27K for Free) means longer, more coherent conversations.
For developers who use the API directly, this update does not change API pricing or model availability — those remain separate from the ChatGPT consumer experience.
Does the reasoning slider affect how much I pay?
No. The slider does not change your subscription price. Higher reasoning levels may consume more of your usage limits (rate limits vary by plan), but the monthly cost stays the same regardless of which level you choose.
Can I switch back to Luna if I have a Plus subscription?
Luna is available on all plans, including Plus and Pro. If you prefer faster, lighter responses for simple tasks, you can use Luna while keeping Sol for more demanding work. The model picker lets you choose.
Is GPT-5.6 Sol available via the API at the same price?
No. This update is specific to the ChatGPT consumer experience. API pricing and model availability are unchanged. GPT-5.6 Sol via the API remains the same version as before — the Chat-optimised tuning applies only inside ChatGPT.
