GPT-5.6 for Professionals: What Changed, Who Should Care, and What to Actually Do

August 6 update — Sol improvements + free-tier model change. OpenAI improved GPT-5.6 Sol: responses are now more direct and concise, and internal testing shows 68% fewer factual errors compared to GPT-5.5 Instant on finance, medical, and legal prompts. Plus and Pro users get a new reasoning slider (five levels of depth, replacing the coarser effort controls). Free and Go users now receive GPT-5.6 Luna as their default model in standard chat — replacing GPT-5.5 — with unlimited text chats and a new "Think" button that lets Luna reason longer on harder questions. Updated plan access is reflected throughout this post.

July 30 update — Price cuts. OpenAI cut API prices for two GPT-5.6 models effective that date. For standard requests with up to 272K input tokens, Luna drops 80% to $0.20/$1.20 per million input/output tokens (was $1/$6). Terra drops 20% to $2/$12 (was $2.50/$15). Sol is unchanged at $5/$30. OpenAI's current Codex credit table also reflects the cheaper Terra and Luna rates. Updated prices are reflected throughout this post.

TL;DR. GPT-5.6 — the Sol/Terra/Luna family — has been broadly available since July 9, 2026. As of August 6, Sol is more accurate and concise, Plus/Pro have a reasoning slider, and free users moved from GPT-5.5 to GPT-5.6 Luna with unlimited text chats. For professionals, the real changes are better accuracy in high-stakes domains (finance, law, medicine), a finer effort dial, and an expanded free tier that trades raw power for unlimited access. One genuine caution: in Full-Access agentic mode, Sol has deleted users' files — keep Auto-review on.

OpenAI's GPT-5.6 launch was messy enough — a government-restricted preview, a 13-day wait, then a broad release — that most professionals reasonably tuned it out. This post covers the full picture from launch through the August 6 improvements: what actually changed for the work you do, who should care, and what (if anything) to change.

If you want the blow-by-blow news — the restricted June 26 preview, the 50-year math proof, the file-deletion incidents as they unfolded — that's in our GPT-5.6 news post. This one is the professional takeaways.

What GPT-5.6 is

GPT-5.6 is a three-model generation, broadly available since July 9, 2026:

Model	Best for	Standard API price per 1M tokens (input / cached input / output)
Sol	Complex agentic work, coding, research	$5 / $0.50 / $30
Terra	Balanced everyday professional use	$2 / $0.20 / $12 (20% cut Jul 30)
Luna	Fast, high-volume, low-cost tasks	$0.20 / $0.02 / $1.20 (80% cut Jul 30)

Those are the current standard, short-context rates. Requests with more than 272K input tokens are billed at 2× the input rate and 1.5× the output rate for the whole request. Cache writes cost 1.25× uncached input; Batch and Flex are half the standard rates; Fast mode is twice standard. All three models now list a 1.05M-token context window, 922K maximum input, 128K maximum output, and a February 16, 2026 knowledge cutoff in OpenAI's model catalog.

For API integrations, the model IDs are gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna; the gpt-5.6 alias routes to Sol. Codex subscribers can select Sol, Terra, or Luna directly. In regular ChatGPT, the product exposes defaults and effort controls rather than raw model IDs.

Who gets what, by plan (full plan breakdown in ChatGPT pricing explained):

• Free — GPT-5.6 Luna in standard chat (unlimited text chats); Think button for harder questions (updated August 6, was GPT-5.5). GPT-5.6 Terra in Codex and ChatGPT Work.
• Go ($8/mo) — GPT-5.6 Luna in standard chat (unlimited text chats); Think button (updated August 6, was GPT-5.5). GPT-5.6 Terra in Codex and ChatGPT Work.
• Plus / Business ($20–25/mo) — Updated GPT-5.6 Sol in standard ChatGPT, with the new reasoning slider (five depth levels). Selectable Sol, Terra, and Luna in Codex. This is the tier where Sol's improved accuracy reaches most professionals.
• Pro / Enterprise ($100–200/mo or custom for Enterprise) — Higher usage plus eligible access to Ultra mode, which delegates suitable work to parallel sub-agents, and the Sol Pro intelligence level.

GPT-5.6 is also the preferred model in Microsoft 365 Copilot — if your organization runs Copilot in Outlook, Teams, Word, or Excel, you got the upgrade automatically with no subscription change.

What genuinely changed (and what didn't)

August 6: Sol is measurably more accurate. OpenAI tuned Sol for directness and reliability. Responses now cut unnecessary detail and excess formatting, delivering tighter answers for simple questions and fuller responses only where complexity warrants it. In an internal evaluation using finance, medicine, and law prompts, responses containing at least one factual error fell about 68% compared to GPT-5.5 Instant. The improvement is most pronounced on questions involving dates, numbers, rules, sources, and assumptions — exactly the domains where professionals most need to trust model output.

August 6: The reasoning slider replaces coarser effort controls. Plus and Pro subscribers now have a five-level slider — from quick answers to deep analysis — replacing the earlier Medium/High/Extra High controls. The principle is the same (spend more reasoning on harder work, less on routine tasks) but with finer calibration.

August 6: Free and Go users moved to GPT-5.6 Luna. Previously, free users had GPT-5.5 in standard chat and only accessed GPT-5.6 in Codex and Work. Now GPT-5.6 Luna is their standard-chat default, with unlimited text chats. The Think button lets Luna reason longer on a specific question. Trade-off: Luna is the lightest GPT-5.6 tier — faster and cheaper, but not as accurate as Sol, particularly on high-stakes professional tasks. If factual accuracy on finance, legal, or medical questions matters, Sol (Plus plan) is the right tier.

Sustained multi-step work remains the core gain. Like Claude's recent releases, the improvement that matters day-to-day isn't a benchmark point — it's follow-through. Sol scores 88.8% on Terminal-Bench 2.1 (agentic/coding tasks), and the practical translation is fewer stalls in long tasks: research-then-draft workflows, multi-document analysis, spreadsheet builds that previously died at step six. If your prompts are single questions, you will notice much less difference.

ChatGPT Work is still the bigger product change. Launched July 9, ChatGPT Work takes an outcome and returns a finished artifact — a report, deck, or spreadsheet assembled from your connected apps — rather than a chat reply. Free and Go get Terra; Plus and higher plans can choose across the family. Usage is metered by task complexity, so pilot before scheduling recurring tasks. Its closest competitor is Claude Cowork — see Codex vs Claude Cowork for how the ecosystems differ.

July 30 pricing. OpenAI cut Luna's standard short-context API price by 80% (to $0.20/$1.20 per million input/output tokens) and Terra's by 20% (to $2/$12). Sol is unchanged. The current Codex credit schedule prices Luna at one-tenth of Terra and Terra at 40% of Sol for the same token volume, which makes the smaller models stretch a subscription allowance much further.

The one genuine caution: Full-Access mode

Two incidents after launch are worth every professional's attention, because they're about configuration, not exotic edge cases. Running in Full-Access mode — the setting that gives the model unsandboxed system control — GPT-5.6 Sol deleted nearly all files on one user's Mac and wiped another user's production database. OpenAI confirmed the root cause on July 16: the model mishandles the $HOME environment variable when setting up temporary directories, and can rm -rf the wrong thing.

If you use ChatGPT only through the chat interface, this does not affect you. If you use ChatGPT Work or Codex agentically:

1. Keep Auto-review enabled. It blocks high-risk actions before they execute.
2. Never point Full-Access at anything you can't restore. Sandboxes and copies exist for this.
3. Back up before long agentic sessions.

The general lesson applies to every vendor's agent, including Claude's: give an AI agent the file permissions you'd give a brand-new contractor, not an administrator.

Who should care, concretely

You're on ChatGPT Plus already: you have the improved Sol now. The August 6 accuracy gains are most useful for any work where you were double-checking model output on factual questions — financial summaries, medical reference lookups, legal rule checks. Try the reasoning slider: use quick for routine tasks, deep for the work where errors are costly. The single highest-value habit change from the July launch still applies: deliberate effort-matching for long multi-step tasks.

You're on Free or Go: your standard chat now uses GPT-5.6 Luna with unlimited text messages — a meaningful change if you were hitting message limits with GPT-5.5. Use the Think button on questions where you'd normally verify the answer externally. For work requiring Sol's improved accuracy (finance, legal, medical) or the full reasoning slider, Plus is where that access starts.

You're a Claude user wondering about switching: the August 6 Sol improvements narrow the accuracy gap in financial and legal domains, which is worth noting. But capability alone doesn't justify switching — Sol and Anthropic's top models trade wins, and Claude Sonnet 5 covers most professional work at the $20 tier. The real asymmetry in 2026 is frontier access economics: OpenAI includes its flagship in Plus, while Anthropic's Fable 5 moved to Max-only with Pro users pushed to API pricing as of July 20. If your work needs frontier-tier capability on a $20 budget, that currently favors OpenAI. For the fuller three-way picture, see Anthropic vs OpenAI vs Google.

You're an API user or run automations: the August 6 Sol improvements apply to API calls as well — if you have pipelines where factual accuracy on financial, medical, or legal content matters, it's worth re-evaluating quality with the updated model before spending time on prompting workarounds. Terra at $2/$12 remains the workhorse for balanced everyday tasks. Luna at $0.20/$1.20 is now dramatically cheaper and is OpenAI's own choice for free-tier standard chat — a practical default for high-volume automation where quality doesn't need to be flagship-level.

You manage a team on M365 Copilot: your users are already on GPT-5.6 Sol, and the August 6 accuracy improvements apply. Worth a note to staff that Copilot output quality may have improved on fact-heavy tasks — and that any internal guidance written against GPT-5.5 behavior may need a refresh.

The bottom line

GPT-5.6 is a solid generational step whose headline improvements — sustained agentic work, effort controls, a bundled task agent — mostly reward people who use AI for multi-step professional work rather than quick questions. The August 6 update sharpens Sol's accuracy meaningfully (68% fewer factual errors in finance/law/medicine) and moves the entire user base onto the GPT-5.6 family. Free users trade GPT-5.5 for unlimited Luna; paid users get a more reliable Sol and finer control. The launch pattern of 2026 remains clear: OpenAI and Anthropic ship comparable capability on similar timelines, and the durable differences are access economics, usage limits, and which ecosystem your work already lives in.

Sources

• OpenAI: Improving GPT-5.6 Sol in ChatGPT — and expanding access to GPT-5.6 Luna for free users (August 6, 2026 announcement)
• 9to5Mac: OpenAI updating ChatGPT with a smarter GPT-5.6 Sol and unlimited free chats (August 6, 2026)
• The Decoder: OpenAI improves GPT-5.6 Sol in ChatGPT and restricts free users to its weakest model (August 6, 2026)
• Neowin: OpenAI announces unlimited GPT-5.6 Luna access for ChatGPT free users (August 6, 2026)
• Axios: OpenAI makes major upgrades for free and paid ChatGPT users (August 6, 2026)
• OpenAI: GPT-5.6: Frontier intelligence that scales with your ambition
• OpenAI: API pricing
• OpenAI: GPT-5.6 Sol model · Terra · Luna
• OpenAI: Previewing GPT-5.6 Sol
• OpenAI: GPT-5.6 is now the preferred model in Microsoft 365 Copilot
• OpenAI Help Center: Model release notes
• ZDNet: OpenAI's GPT-5.6 and ChatGPT Work aim to beat Anthropic on price, speed, and productivity
• Forbes: OpenAI Launches GPT-5.6 and ChatGPT Work AI Agent
• The Register: OpenAI admits GPT-5.6 occasionally deletes files — but it's an 'honest mistake'
• TechCrunch: OpenAI's new flagship model deletes files on its own, people keep warning
• The Decoder: OpenAI goes full China pricing mode with an 80 percent cut to its most affordable GPT-5.6 model (July 30, 2026 price cuts)
• 9to5Mac: OpenAI makes two GPT-5.6 models cheaper, expanding usage in ChatGPT (subscription plan implications)
• OpenAI: Advancing the price-performance frontier with GPT-5.6 (July 30, 2026 price cuts)

Plan access, model behavior, and prices are an August 6, 2026 snapshot — OpenAI adjusts these frequently, so confirm against the live product before deciding. This article is informational, not purchasing advice.