- Should you update your business workflows with Sol and Luna now?
- What really changed with GPT-5.6 Sol on August 6, 2026?
- GPT-5.6 Luna for free: what do you actually get without paying?
- How much does it cost to use Sol via API in business processes now?
- Real-world scenarios: what changes for three types of Italian companies?
- Why are 76% of Italian SMEs still out of the AI game?
- What are the real limits of Sol and Luna that nobody mentions?
- 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 can you tell if ChatGPT is really 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 no entry cost.
- Sol's API price dropped by 20-33% in August 2026: integrating ChatGPT into business processes costs less.
- 76% of Italian SMEs have yet to invest in AI: this update lowers the barrier, but you need a strategy so you don't waste the opportunity.
Updated October 3, 2026
GPT-5.6 Sol has cut incorrect answers 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 anyone using ChatGPT at work.
Should you update your business workflows with Sol and Luna now?
It depends on what you already do with ChatGPT. Sol is the right choice if you need verifiable, repeatable output; Luna is the ideal starting point if you haven't begun yet.
On August 6, 2026, OpenAI published an official update covering two separate models: GPT-5.6 Sol, already available to subscribers, becomes more precise and less error-prone; GPT-5.6 Luna, the lighter and faster variant, becomes free with unlimited conversations for everyday use.
Two separate moves, one shared goal: widening the base of people who use ChatGPT seriously. Sol aims at professional quality, while Luna lowers the entry threshold. 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 produce content, analyses or reports on a regular basis | Sol (paid plan) | Greater accuracy and consistency on repeated tasks |
| You want to test generative AI without spending anything | Luna (free plan) | Unlimited conversations, zero entry cost |
| You integrate ChatGPT via API into your business systems | Sol via API | Price reduced by 20-33% since August 2026 |
| You have a team to train on AI before structuring processes | Luna as an onboarding tool | No cost, ideal for the exploratory phase |
| You operate in regulated sectors (legal, medical, financial) | Sol Enterprise with contractual safeguards | 68% fewer factual errors compared to the previous model |
What really changed with GPT-5.6 Sol on August 6, 2026?
According to OpenAI's internal evaluation, Sol cut incorrect answers by 68% on financial, medical and legal prompts compared to GPT-5.5 Instant. Consistency across similar sessions has improved in measurable ways.
GPT-5.6 Sol was already the go-to model for ChatGPT Plus, Team and Enterprise subscribers. Earlier versions showed three concrete problems: inconsistent answers on complex prompts, occasional factual errors on technical data, and variability between similar sessions. Anyone using Sol to produce reports or analyses knew the result could change from one session to the next.
The August 2026 update addresses all three. According to OpenAI, Sol produced 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) coding benchmark, Sol with maximum reasoning scored 80, beating Claude Fable 5 by 2.8 points while using less than half the output tokens at about a third of the cost.
If you handle high volumes of content or work in fields where a mistake has real consequences, these numbers aren't just marketing: they mean less time spent checking and correcting the model's output.
GPT-5.6 Luna for free: what do you actually get without paying?
Luna is the variant optimized for speed and efficiency. Since August 2026 it's available for free with unlimited conversations for everyday use, but it doesn't replace Sol for complex professional tasks.
Until the August 2026 update, Luna was reserved for subscribers. Now anyone can use it with no daily conversation limits. According to 01net.it, Luna produced 62% fewer incorrect answers than GPT-5.5 Instant in the same internal OpenAI evaluation.
The difference between Sol and Luna isn't just price: it's usage profile. Luna is faster and uses fewer computational resources. It suits standard conversations, quick searches, draft texts and everyday browsing support. Sol is built for tasks that need deep reasoning, consistency across long texts and verifiable output.
| Feature | GPT-5.6 Sol | GPT-5.6 Luna |
|---|---|---|
| Access | Paid plan (Plus, Team, Enterprise) | Free, unlimited conversations |
| Error reduction vs GPT-5.5 Instant | -68% | -62% |
| Optimized for | Complex tasks, repeatable output | Standard conversations, speed |
| API use | Yes | Limited |
| Ideal for | Structured professional workflows | Exploration, training, everyday use |
How much does it cost to use Sol via API in business processes now?
On August 21, 2026, OpenAI cut Sol's API pricing: input -20%, output -33%. The promotional price is valid through at least November 21, 2026.
For companies that integrate ChatGPT into their systems via API (a programming interface that lets two pieces of software talk to each other), the cost per million tokens, the unit used to measure processed text, has dropped significantly. According to the official OpenAI community, the input price went from $5 to $4 per million tokens (-20%), and the output price from $30 to $20 (-33%).
In practical terms: a company that processes large volumes of text, such as product sheets, reports and automated replies, sees its operating costs fall without changing anything in its integration. The savings are automatic for anyone already using Sol via API.
If you're considering building automated workflows, for example to generate product descriptions, summarize documents or answer standard requests, you now have one more financial argument in favor. We explore AI integration in business processes 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 illustrative scenarios show how to calculate the real benefit.
Illustrative examples: numbers are meant to show the reasoning, they are not real customer data.
| Company type | Current ChatGPT use | What changes with Sol/Luna | Main benefit |
|---|---|---|---|
| E-commerce with 5,000 product sheets | Description generation via API 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 summaries and draft replies | Sol cuts factual errors by 68% on legal and financial prompts | Fewer manual checks, more reliable output |
| Manufacturing SME with no in-house AI | No structured use | Free Luna lets you get started without a dedicated budget | Zero-cost team training, in-house AI culture |
For SMEs that haven't started yet, Luna is a concrete starting point. For those already using ChatGPT systematically, the updated Sol removes the main barrier to adoption: inconsistent output. This is exactly the topic we covered in ChatGPT for Small Business and Italian SMEs.
Why are 76% of Italian SMEs still not using AI?
According to the Politecnico di Milano Observatory, three quarters of Italian SMEs have neither invested in AI nor plan to. The problem isn't cost: it's the lack of an entry strategy.
The Italian market data is clear. According to the Politecnico di Milano's Digital Innovation in SMEs Observatory (2025-2026 survey, published in May 2026), 76% of Italian SMEs have neither invested in artificial intelligence nor plan to. Only 7% have started structured AI training.
The contrast with large companies is stark. According to ISTAT data (Enterprises and ICT 2025), 53.1% of large Italian companies already use AI, up from 32.5% the year before. Among SMEs, adoption is stuck at 15.7%.
ChatGPT passed 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 under three years from launch. Italian SMEs that stay out aren't just missing a tool: they're losing ground to competitors who already use it.
What are the real limits of Sol and Luna that nobody mentions?
Improved accuracy doesn't mean infallibility. There are still critical areas every company should know about before building processes that depend on ChatGPT.
The improvements claimed by OpenAI are real, but they need context. The 68% error reduction figure comes from internal testing and covers specific categories of prompts. In real business use, variability may differ. Our practical advice: test Sol on concrete tasks from your own workflow before redesigning established processes.
Three areas remain critical even after the update:
| Critical area | Current situation | How to handle it |
|---|---|---|
| Context in very long conversations | Still a weak point for Sol on very long texts | Break tasks into shorter, structured sessions |
| Integration with business systems (ERP, CRM) | Requires custom configurations or dedicated APIs | Consider an Enterprise plan or a tailored integration |
| Privacy and GDPR | Sensitive company data should not be entered without contractual safeguards | Use the Enterprise plan with a Data Processing Agreement, or on-premise solutions |
Privacy is especially relevant for European companies. The GDPR sets precise limits on using sensitive company data on external AI platforms. Before using ChatGPT to process customer data, contracts or confidential information, you need to check the contractual terms of your chosen plan. We explored this topic in the context of ChatGPT's persistent memory for B2B SMEs.
How do you use Sol in marketing and content workflows?
A more accurate and consistent Sol improves reliability in every task that requires repeatable output: copywriting, analysis, editorial production, automated replies.
For marketing managers, the update has immediate practical effects. Inconsistent output was the main barrier to systematic adoption in professional workflows. With a more consistent Sol, some tasks become more reliable to automate or semi-automate.
| Marketing activity | Benefit with updated Sol | Operational note |
|---|---|---|
| Creative briefs and campaign copy generation | More consistent output across different sessions | Test on 10-20 real prompts before automating |
| Competitor analysis and report summaries | Fewer factual errors on numerical data and quotes | Always verify critical data against primary sources |
| Editorial production for blogs and SEO | Better stylistic consistency across long articles | Define a system prompt with tone and guidelines |
| Customer-facing reply automation | More consistent answers to similar questions | Use for standard FAQs, not complex cases |
| Internal team training | Free Luna as an onboarding tool | Set up hands-on exercises before moving to Sol |
ChatGPT's evolution is part of a broader shift toward segmentation: free, professional, enterprise. Each tier has different models, limits and guarantees. To see how it stacks up against competitors, it's worth reading our analysis of the new Gemini 2026 models and what they mean for Italian SMEs. The arrival of GPT-6 Astra also points the same way: models that are increasingly specialized for specific verticals.
How to get started: concrete steps to begin with Sol or Luna
The most effective path starts with free exploration using Luna, identifies high-volume tasks, then considers Sol for processes that require verifiable outputs.
- 1 · Manager or teamStart with free LunaSign up for a free ChatGPT account and use Luna for 2-3 weeks on real everyday tasks: summaries, drafts, quick research.
- 2 · ManagerMap your high-volume tasksIdentify the activities your team repeats every week that require structured text: reports, product sheets, standard replies.
- 3 · IT or digital managerTest Sol on critical tasksActivate 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 teamCheck the GDPR termsBefore entering sensitive company data, review the terms of the plan you chose. For confidential data, consider the Enterprise plan with a Data Processing Agreement.
- 5 · ManagerBuild the workflow and train the teamDefine system prompts, editorial guidelines and an output verification process. Train your team with hands-on exercises on real cases.
- 6 · ManagerMeasure and optimizeTrack time saved, the error rate in outputs and team satisfaction. Review the workflow every quarter.
What are the most common mistakes when adopting ChatGPT in a company?
The three most frequent mistakes 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 check. A 68% reduction in errors means Sol makes fewer mistakes, not that it never makes them. For published content or official documents, review is still necessary.
The second mistake is skipping training. According to the AI4Innovation Observatory of Politecnico di Milano, 73% of the Italian companies analyzed name employee access to generative tools as their main AI investment, based on data published by Res Group. But giving access without training leads to mediocre results and frustrated teams.
The third mistake is failing to distinguish between exploratory and professional use. Luna is great for getting started, but building critical business processes on a free model without contractual guarantees is a risk many companies underestimate. The guide to business workflows with ChatGPT Voice Mode goes deeper into how to structure professional use sustainably.
How can you tell if ChatGPT is really improving your processes?
Three concrete indicators: output review time, error rate in published content, and hours saved each week on repetitive tasks.
| Indicator | How to measure it | Positive signal |
|---|---|---|
| AI output review time | Time the average editing time by 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 | Gradually decreases with more refined prompts |
| Hours saved on repetitive tasks | Compare time before and after adoption 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 you use the API) | Calculate the cost per document or response generated | Drops with prompt optimization and API price cuts |
So, is it worth it for your company?
Worth it if you already use ChatGPT systematically and need more reliable outputs: the updated Sol cuts errors by 68% and costs less via API. The time to structure your workflows is now.
A middle path: you're evaluating AI but don't have a defined process yet. Start with free Luna to train the team, then consider Sol once you've identified the high-volume tasks that need professional-grade quality.
Not yet if you have no internal process for verifying AI outputs and aren't ready to train your team: adopting ChatGPT without these foundations leads to mediocre results and needless risk. Start with training.
How we see it at SHM
For us, the Sol and Luna update confirms something we often repeat: AI isn't a magic formula, it's a tool that only works with a solid human process behind it. A more accurate Sol is good news, but the real competitive advantage doesn't come from the model. It comes from how you use it, who uses it and what you do with the outputs.
We believe that integrating AI into a company means, first of all, understanding which processes are worth automating and which aren't. Free Luna is a real opportunity for Italian SMEs that haven't started yet: it lowers the barrier to entry without requiring investment. But the leap from "occasional use" to "structural advantage" takes a strategy, not just an account.
With our strategic AI integration and AI training for companies services, we support Italian SMEs along this path: from identifying the tasks suited to AI, to structuring workflows, to training the team. If you want to know where to start, our AI and process automation hub collects all our practical insights. You can also contact us directly to talk through your specific case.
Frequently Asked Questions
Yes. Luna is optimized for speed and standard conversations. It isn't suited to complex tasks that call for deep reasoning, consistency across very long texts, or systematically verifiable output. For structured professional use, Sol remains the right choice.
Not without the right contractual guarantees. The GDPR sets precise constraints on using sensitive data on 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 information.
The price has dropped to $4 per million input tokens (-20%) and $20 per million output tokens (-33%). According to OpenAI's official announcement, the promotional price is valid until at least November 21, 2026.
Start with free Luna for 2-3 weeks on real tasks. Then test Sol on the same prompts and measure the review time needed. If Luna's outputs need a lot of corrections on critical tasks, Sol justifies the subscription cost.
On the Artificial Analysis Coding Agent Index v1.1 benchmark, Sol at maximum reasoning scored 80 points, beating Claude Fable 5 (77.2) while using less than half the output tokens at about one third of the cost. Benchmarks vary by task type: it's always worth testing on your own specific use cases.
According to the Politecnico di Milano Observatory (May 2026), the main problem isn't cost but the lack of an entry strategy and internal training. Only 7% of SMEs have launched structured AI programs. Free Luna lowers the cost barrier, but you still need a methodical approach to get concrete results.
The model improves every month: the competitive advantage doesn't go to whoever adopts it first, but to whoever 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.
