- Is it worth upgrading company workflows with Sol and Luna right now?
- What really changed with GPT-5.6 Sol on August 6, 2026?
- GPT-5.6 Luna free: what do you really get without paying?
- How much does it cost now to use Sol via API in business processes?
- Real-world scenarios: what changes for three types of Italian companies?
- Why is 76% of Italian SMEs still left out of AI?
- What are the real limits of Sol and Luna that nobody talks about?
- How do you use Sol in marketing and content workflows?
- How to get started: concrete steps to begin with Sol or Luna?
- What are the most common mistakes when adopting ChatGPT in a company?
- How to figure out if ChatGPT is actually improving your processes?
- So, is it worth it for your company?
- How we see it at SHM
- Frequently Asked Questions
- GPT-5.6 Sol is 68% more accurate than GPT-5.5 Instant on financial, medical, and legal prompts: it becomes more reliable for repeatable professional workflows.
- GPT-5.6 Luna is now free with unlimited conversations: SMEs can test generative AI with zero entry costs.
- Sol's API price dropped by 20-33% in August 2026: integrating ChatGPT into business processes costs less.
- 76% of Italian SMEs have not yet invested in AI: this update lowers the barrier, but a strategy is needed to avoid wasting the opportunity.
Updated October 3, 2026
GPT-5.6 Sol has reduced incorrect responses by 68% compared to the previous model. On the same day, OpenAI made GPT-5.6 Luna free for everyone: a double update that changes the game for those who use ChatGPT in business.
Is it worth upgrading company workflows with Sol and Luna right now?
It depends on what you already do with ChatGPT. Sol is the right choice for those who need verifiable and repeatable output; Luna is the ideal starting point for those who haven't started yet.
On August 6, 2026, OpenAI has released an official update on two distinct models: GPT-5.6 Sol, already available for subscribers, becomes more accurate and less prone to errors; GPT-5.6 Luna, the lighter and faster variant, becomes free with unlimited conversations for daily use.
Two separate initiatives, but with a common goal: to broaden the base of serious ChatGPT users. Sol aims for professional quality, Luna lowers the entry barrier. For an Italian company, the practical question is: which of the two does my team really need?
| Your situation | Recommended model | why |
|---|---|---|
| You use ChatGPT to systematically produce content, analyses, or reports | Sol (paid plan) | Greater accuracy and consistency on repeated tasks |
| You want to test generative AI without spending a dime | Luna (free plan) | Unlimited conversations, zero entry costs |
| Integrate ChatGPT via API into your business systems | Sol via API | Price reduced by 20-33% starting August 2026 |
| You have a team to train on AI before structuring processes | Luna as an onboarding tool | Free of charge, ideal for the exploratory phase |
| You operate in regulated industries (legal, medical, financial) | Enterprise solution with contractual guarantees | 68% reduction in factual errors compared to the previous model |
What really changed with GPT-5.6 Sol on August 6, 2026?
Sol has reduced incorrect responses on financial, medical, and legal prompts by 68% compared to GPT-5.5 Instant, according to OpenAI's internal evaluation. Consistency between similar sessions has improved measurably.
GPT-5.6 Sol was already the benchmark model for ChatGPT Plus, Team, and Enterprise subscribers. Previous versions showed three concrete problems: inconsistent responses on complex prompts, occasional factual errors on technical data, and variability between similar sessions. Those who used Sol to produce reports or analyses knew that the result could change from one session to the next.
The August 2026 update addresses all three points. According to OpenAI , Sol scored 68% fewer responses containing at least one factual error compared to GPT-5.5 Instant, in an evaluation on financial, medical, and legal prompts. On the Artificial Analysis Coding Agent Index (v1.1) code benchmark, Sol with maximum reasoning scored 80, beating Claude Fable 5 by 2.8 points while using less than half the output tokens and at about one-third of the cost.
For those who handle high volumes of content or work in sectors where an error has concrete consequences, these numbers are not just marketing: they mean less time spent verifying and correcting model outputs.
GPT-5.6 Luna free: what do you really get without paying?
Luna is the optimized variant for speed and efficiency. From August 2026 it will be available for free with unlimited conversations for daily use, but it does not replace Sol for complex professional tasks.
Until the August 2026 update, Luna was reserved for subscribers. Now anyone can use it without daily conversation limits. According to 01net.it , Luna got 62% fewer wrong answers than GPT-5.5 Instant in the same internal test by OpenAI.
The difference between Sol and Luna isn't just about price: it's about how you use them. Luna is faster and uses fewer computing resources. It's great for standard chats, quick searches, writing drafts, and everyday browsing help. Sol is built for tasks that need deep thinking, consistency in long texts, and outputs you can check.
| Feature | GPT-5.6 Sol | GPT-5.6 Luna |
|---|---|---|
| Access | Paid plan (Plus, Team, Enterprise) | Free, unlimited chats |
| Error reduction vs GPT-5.5 Instant | -68% | -62% |
| Optimized for | Complex tasks, repeatable outputs | Standard chats, speed |
| API usage | Yes | Limited |
| Ideal for | Structured professional workflows | Exploration, training, daily use |
How much does it cost now to use Sol via API in business processes?
On August 21, 2026, OpenAI cut Sol's API price: input -20%, output -33%. The promotional price is valid at least until November 21, 2026.
For companies integrating ChatGPT into their systems via API (Application Programming Interface, which allows two software programs to communicate with each other), the cost per million tokens—the unit of measurement for processed text—has significantly decreased. According to the official OpenAI community , the input price dropped from $5 to $4 per million tokens (-20%), and the output price from $30 to $20 (-33%).
In practical terms: a company processing large volumes of text—product descriptions, reports, automated responses—sees its operating costs reduced without changing anything in its integration. Savings are automatic for those already using Sol via API.
Anyone considering setting up automated workflows — like generating product descriptions, summarizing documents, or handling standard requests — now has an extra economic reason to do so. The topic of AI integration in business processes is explored in depth in our article on AI agents and business workflows in 2026 .
Real-world scenarios: what changes for three types of Italian companies?
The impact of Sol and Luna depends on your industry and workload. Three examples show how to figure out your real benefit.
Illustrative examples: the numbers serve to show the reasoning, they are not real customer data.
| Company type | Current use of ChatGPT | What changes with Sol/Luna | Main benefit |
|---|---|---|---|
| E-commerce with 5,000 product pages | API description generation with GPT-5.5 | API output cost -33%; fewer errors to fix manually | Savings on API costs and review time |
| Professional firm (legal/tax) | Document summarization and reply drafts | Sol reduces factual errors by 68% on legal and financial prompts | Fewer manual checks, more reliable outputs |
| Manufacturing SME without in-house AI | No structured use | Free Luna lets you get started with zero dedicated budget | Team training at zero cost, internal AI culture |
For SMBs that haven't started yet, Luna is a practical starting point. For those already using ChatGPT systematically, the updated Sol reduces the main obstacle to adoption: output variability. This is exactly the topic we discussed when talking about ChatGPT for Small Business and Italian SMBs .
Why is 76% of Italian SMEs still left out of AI?
According to the Politecnico di Milano Observatory, three-quarters of Italian SMEs have not invested or do not plan to invest in AI. The problem is not the cost: it's the lack of an entry strategy.
Italian market data is clear. According to the Digital Innovation Observatory in SMEs by the Polytechnic University of Milan (survey 2025-2026, published in May 2026), 76% of Italian SMEs have not invested or do not plan to invest in artificial intelligence. Only 7% have started structured training on AI.
The contrast with large enterprises is stark. According to ISTAT data (Businesses and ICT 2025) , 53.1% of large Italian enterprises already use AI, up from 32.5% the previous year. In SMEs, adoption stands at 15.7%.
ChatGPT surpassed one billion monthly active users in June 2026, according to AI4Business which cites Sensor Tower estimates reported by Reuters. It is the first app in history to reach this milestone in less than three years since launch. Italian SMEs that stay out are not just missing out on a tool: they are losing ground to competitors who are already using it.
What are the real limits of Sol and Luna that nobody talks about?
Improved accuracy doesn't mean infallibility. There are still critical areas that every company needs to know about before structuring processes dependent on ChatGPT.
The improvements claimed by OpenAI are real, but need to be put into perspective. The 68% error reduction claim is internal and applies to specific prompt categories. In real-world business settings, variability might look different. The practical advice is to test Sol on actual tasks in your workflow before redesigning established processes.
Three areas remain critical even after the update:
| Critical area | Current situation | How to manage it |
|---|---|---|
| Context on very long conversations | Still a weak spot for Sol on very long texts | Break tasks down into shorter, structured sessions |
| Integration with corporate systems (ERP, CRM) | Requires custom setups or dedicated APIs | Evaluate an Enterprise plan or a custom integration |
| Privacy and GDPR | Sensitive corporate data should not be entered without contractual guarantees | Use the Enterprise plan with a Data Processing Agreement or on-premise solutions |
The privacy issue is particularly relevant for European companies. GDPR imposes strict rules on using sensitive business data in external AI platforms. Before using ChatGPT to process customer data, contracts, or confidential info, you need to check the contract terms of your chosen plan. We explored this topic in depth in the context of ChatGPT persistent memory for B2B SMBs .
How do you use Sol in marketing and content workflows?
A more accurate and consistent Sol improves reliability in all activities that require repeatable output: copywriting, analysis, editorial production, response automation.
For marketing managers, the update has immediate practical implications. Output variability was the main obstacle to systematic adoption in professional workflows. With a more consistent Sol, certain tasks become much more reliable to automate or semi-automate.
| Marketing activities | Advantage with updated Sol | Operational note |
|---|---|---|
| Generation of creative briefs and copy for campaigns | More consistent output across different sessions | Test on 10-20 real prompts before automating |
| Competitor analysis and report synthesis | Fewer factual errors on numerical data and citations | Always verify critical data with primary sources |
| Content production for blogs and SEO | Greater stylistic consistency on long articles | Define a system prompt with tone and guidelines |
| Automated customer-facing responses | More consistent answers on similar questions | Reserve for standard FAQs, not complex cases |
| Internal team training | Free month as an onboarding tool | Set up practical exercises before moving on to Sol |
The evolution of ChatGPT is part of a broader context of market segmentation: free, professional, enterprise. Each tier has different models, limits, and guarantees. To understand how it stacks up against competitors, it's worth reading the analysis on new Gemini 2026 models and the implications for Italian SMEs . Also the arrival of GPT-6 Astra shows the direction: increasingly specialized models for specific verticals.
How to get started: concrete steps to begin with Sol or Luna?
The most effective path starts with free exploration with Luna, identifies high-volume tasks, then evaluates Sol for processes requiring verifiable output.
- 1 · Manager or team Get started with free Luna Sign up for a free ChatGPT account and use Luna for 2-3 weeks on real daily tasks: summaries, drafts, quick searches.
- 2 · Manager Map high-volume tasks Figure out the tasks your team repeats every week that need structured text: reports, product descriptions, canned responses.
- 3 · IT or digital manager Test Sol on critical tasks Activate a Plus or Team plan and compare Sol's outputs with Luna's on the same prompts. Measure the review time needed.
- 4 · Legal or compliance team Check GDPR conditions Before entering sensitive company data, check the terms of the chosen plan. For confidential data, consider the Enterprise plan with a Data Processing Agreement.
- 5 · Manager Set up the workflow and train the team Define system prompts, editorial guidelines, and an output verification process. Train the team with hands-on exercises on real cases.
- 6 · Manager Measure and optimize Keep track of time saved, error rates in outputs, and team happiness. Check your workflow every three months.
What are the most common mistakes when adopting ChatGPT in a company?
The three most frequent errors are: using ChatGPT without a verification process, not training the team, and entering sensitive data without contractual guarantees.
Even with a more accurate Sol, the most common mistake is treating outputs as final without a human review step. A 68% error reduction means Sol makes fewer mistakes, not that it never makes them. For published content or official documents, review remains necessary.
The second mistake is skipping the training phase. 73% of Italian companies analyzed by the AI4Innovation Observatory of the Politecnico di Milano indicate employee access to generative tools as the main area of investment in AI, according to data published by Res Group . But giving access without training leads to mediocre results and frustrated teams.
The third mistake is not distinguishing between exploratory use and professional use. Luna is great for starting, but structuring critical business processes on a free model without contractual guarantees is a risk that many companies underestimate. The guide to business workflows with ChatGPT Voice Mode explores how to structure professional use in a sustainable way.
How to figure out if ChatGPT is actually improving your processes?
Three concrete indicators: output review time, error rate in published content, and hours saved on repetitive tasks every week.
| Indicator | How to measure it | Positive signal |
|---|---|---|
| Time spent reviewing AI outputs | Time the average editing time per task type | Drops by 20-30% after the first month of structured use |
| Error rate in published content | Count post-publication corrections on AI-generated content | It gradually decreases with more refined prompts |
| Hours saved on repetitive tasks | Compare pre- and post-adoption time on standardized activities | At least 2-3 hours/week per person on high-volume tasks |
| Team satisfaction | Monthly internal survey (1-5 scale) | Average score above 3.5 after the first quarter |
| Cost per output (if using APIs) | Calculate the cost per document or generated response | It drops with prompt optimization and the API price cut |
So, is it worth it for your company?
It is worth it if you already use ChatGPT systematically and need more reliable output: updated Sol reduces errors by 68% and costs less via API. The time to structure workflows is now.
A middle ground you are evaluating AI but don't have a defined process yet: start with free Luna to train the team, then evaluate Sol when you have identified high-volume tasks that require professional quality.
Not yet if you don't have any internal process yet to verify AI outputs and you're not ready to train the team: adopting ChatGPT without these foundations produces mediocre results and unnecessary risks. Start with training.
How we see it at SHM
For us, updating Sol and Luna proves something we always say: AI isn't magic, it's a tool that only works if you've got a solid human process behind it. A more accurate Sol is great news, but your real edge doesn't come from the model itself—it comes from how you use it, who's using it, and what you do with the results.
We believe that integrating AI into a company means first of all understanding which processes are worth automating and which are not. Free Luna is a concrete opportunity for Italian SMEs that haven't started yet: it lowers the barrier to entry without requiring investments. But the jump from “occasional use” to “structural advantage” requires a strategy, not just an account.
With our services of strategic AI integration and of AI training for companies we support Italian SMEs on this journey: from identifying suitable AI tasks, to structuring workflows, to team training. If you want to understand where to start, our hub on AI and process automation collects all practical insights. You can also contact us directly to discuss your specific case.
Frequently Asked Questions
Yes. Luna is optimized for speed and standard chats. It's not suited for complex tasks requiring deep reasoning, consistency across very long texts, or systematically verifiable outputs. For structured professional use, Sol remains the right choice.
Not without the right contractual guarantees. GDPR imposes strict rules on using sensitive data in external AI platforms. For confidential data, you need the Enterprise plan with a Data Processing Agreement or an on-premise solution. Always check the plan's terms before entering critical info.
The cost has dropped to $4 per million input tokens (-20%) and $20 per million output tokens (-33%). The promotional price is valid at least until November 21, 2026, according to official OpenAI communication.
Start with free Luna for 2-3 weeks on real tasks. Then test Sol on the same prompts and measure the revision time needed. If Luna's outputs require many corrections on critical tasks, Sol justifies the subscription cost.
On the Artificial Analysis Coding Agent Index v1.1 benchmark, Sol with maximum reasoning scored 80 points, beating Claude Fable 5 (77.2) while using less than half the output tokens and at about a third of the cost. Benchmarks vary by task type: it is always a good idea to test on your specific use cases.
According to the Politecnico di Milano Observatory (May 2026), the main issue isn't the cost but the lack of an entry strategy and internal training. Only 7% of SMEs have started structured AI pathways. Free Luna lowers the financial barrier, but you still need a methodical approach to get concrete results.
The model improves every month: competitive advantage doesn't come from who adopts it first, but from who builds the right process around it.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
