.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Why Is OpenAI Moving Features to the Free Plan When Everyone Else Is Doing the Opposite
For most software companies, the pricing instinct is familiar: build something valuable, put it behind a paid tier, and use access to the feature as the reason to upgrade. OpenAI has repeatedly run the play in reverse. In May 2024, free ChatGPT gained GPT-4-class intelligence, web access, data analysis, file and image uploads, GPTs and memory. Search subsequently moved from paid access to all users. Deep research later reached Free in a limited form. By 21 August 2026, Free users have GPT-5.6 Luna, unlimited everyday text chats subject to safeguards, web search, data analysis, file uploads, image creation and GPT access.
At first glance, this looks commercially backwards. AI inference costs real money. SaaS companies are under pressure to monetise AI rather than subsidise it. Yet the more closely we examine OpenAI's moves - and what Google, Microsoft, Salesforce and HubSpot are doing - the less contradictory the strategy becomes. Monetizely's position is that OpenAI is not dismantling its paywall. It is moving the paywall from feature access to scarce capacity: stronger reasoning, higher limits, longer context, autonomous work, enterprise controls and ad-free attention. Free is becoming the distribution layer. Paid plans increasingly sell how much work the system can do for you.
Free is becoming OpenAI's distribution layer, while paid plans sell scarce capacity
The easiest way to see the change is to compare what OpenAI charged for access to over time. ChatGPT Plus launched on 1 February 2023 at $20 a month. The paid proposition was largely scarcity: access during peak periods, faster responses and priority access to new features. OpenAI explicitly said at launch that subscription pricing would help support continued free availability.
A dated Wayback capture from 10 January 2024 still shows that recognisable structure: Free at the entry point, Plus at $20 a month, alongside Team and Enterprise. By May 2024, however, several capabilities that would have made a plausible premium bundle on their own were being handed to free users.
The sequence matters more than any one launch.
| Date | What changed | What remained scarce | Primary source |
|---|---|---|---|
| Feb. 2023 | Plus launched at $20/month with peak access, faster responses and earlier feature access. | Reliability, speed and priority. | |
| Jan. 2024 | OpenAI's pricing page showed Free, $20/month Plus, Team and Enterprise. | Higher-tier models, limits and workplace controls. | |
| May 2024 | Free gained GPT-4-class intelligence plus web, data analysis, files, images, GPTs and memory, subject to limits. | Plus received materially higher usage limits. | |
| Dec. 2024-Feb. 2025 | ChatGPT Search expanded from paid users to logged-in free users and then broader public access in supported markets. | Higher overall paid-plan capacity. | |
| Apr. 2025 | Lightweight deep research reached Free with a limited allowance; paid tiers received much larger allowances. | Volume and access to more compute-intensive research. | |
| Aug. 2026 | Free has GPT-5.6 Luna, everyday text chat, search, data analysis, uploads, GPT use and image creation. Paid tiers receive GPT-5.6 Sol reasoning options and higher tool limits. | Premium reasoning, larger allowances, context and advanced work. |
The pattern is clear: OpenAI increasingly lets Free answer the question "Can ChatGPT do this?" and asks paid plans to answer "How much, how hard and how continuously can ChatGPT do this?"
That is a much stronger boundary in a fast-moving AI market. Feature gates depreciate quickly. Once web search, file analysis or image creation becomes expected behaviour across competing assistants, withholding the capability weakens the free product without creating durable pricing power.
Compute does not depreciate in the same way. A long reasoning run still consumes more resources than an everyday chat. An autonomous coding task can run for much longer than a single answer. A business still values larger context windows, faster responses, identity management and contractual controls. As of 21 August 2026, OpenAI's Business page, for example, sells not merely "more AI" but a secure workspace, company connections, spend controls, SAML SSO, administration and higher-capability model access.
There is another reason OpenAI can afford a richer Free tier. Free no longer has to mean economically invisible. In 2026, OpenAI began allowing ads on eligible Free and Go experiences while keeping higher paid tiers ad-free. Free can therefore do three jobs at once: acquire users, establish ChatGPT as the default AI interface, and create an advertising audience where OpenAI chooses to monetise attention.
The commercial logic resembles consumer internet economics more than classic SaaS. You do not cripple Google Search to persuade someone to buy Google Workspace. You make the broadly available product good enough to become habitual, then monetise expensive or business-critical use differently.
The market is converging on included AI plus paid work, not retreating behind feature gates
The premise that "everyone else is doing the opposite" is understandable, but it does not survive contact with current enterprise pricing.
Google made one of the clearest moves in January 2025. Gemini capabilities that had previously required a separate Workspace AI add-on were folded into Workspace Business and Enterprise plans. Google's own example said a Business Standard customer previously paying $32 per user per month for Workspace plus Gemini Business would pay $14 after the change. Jerry Dischler, then President of Cloud Applications, summarised the reasoning succinctly: "AI is foundational to the future of work."
Microsoft, Salesforce and HubSpot are approaching the same destination through metered agent usage rather than simply giving everything away.
| Vendor | Included or free layer | What becomes paid as AI does more work | Pricing signal |
|---|---|---|---|
| Google Workspace | Gemini AI folded into Business and Enterprise Workspace plans in Jan. 2025. | Higher Workspace tiers and enterprise requirements rather than a separate AI surcharge for baseline assistance. | AI becomes part of the suite. |
| Microsoft | Microsoft 365 Copilot Chat is available to eligible users without a separate Copilot Chat licence. | Agents can be metered by messages; Microsoft announced $0.01 per message PAYG and capacity packs of $200 for 25,000 messages in Jan. 2025. | Chat is included; autonomous work is metered. |
| Salesforce | Agentforce Foundations is listed at $0. | Flex Credits are listed at $500 per 100,000 credits, with actions consuming credits. | Free foundation; pay as agents act. |
| HubSpot | AI is integrated into its Customer Platform and plans include credit allowances. | From Apr. 2026, Customer Agent costs 50 credits per resolved conversation and Prospecting Agent 100 credits per recommended lead, equivalent to $0.50 and $1 at HubSpot's stated rates. | Included allowance; pay for completed work. |
The table changes the competitive story. OpenAI is more aggressive about the quality of its zero-price consumer product, but the underlying pricing direction is shared: baseline intelligence is being bundled or subsidised, while higher-value machine work becomes the monetisation point.
Practitioners running these businesses are unusually explicit about the shift. Google Cloud CEO Thomas Kurian told investors in September 2025, "Some people pay us for some of our products by consumption." Microsoft CEO Satya Nadella said in April 2026, "We are evolving our family of Copilots from synchronous assistants to async coworkers."
Salesforce CEO Marc Benioff described Agentforce as bringing "AI, data, apps, and automation with humans to reshape how work gets done." HubSpot Chief Customer Officer Jon Dick went even further in April 2026: "Outcome-based pricing removes that risk. You pay when it works, full stop."
Those statements describe a category moving from charging for intelligence to charging for labour performed by intelligence.
OpenAI's packaging is ahead of the metric it uses to charge
Monetizely's 5-Step Pricing Framework starts with Goals and Segmentation, establishing what the business needs pricing to accomplish and which customers it serves. Packaging then turns those segment needs into distinct offers. Pricing Metric decides the unit against which revenue scales - a seat, usage, transaction, outcome or another measurable unit. Rate Setting determines the actual price attached to that structure. Operationalization makes the promise executable through entitlements, metering, billing, spend controls and customer communication. The sequence matters because a clever rate cannot rescue the wrong package or meter. As developed in Monetizing Agentic AI, the framework is especially useful here because AI causes costs and customer value to vary far more dramatically by task than they did in classic SaaS.
For OpenAI, three steps deserve particular scrutiny.
| 5-Step Framework step | Grade | Monetizely's assessment |
|---|---|---|
| Packaging | A- | Free establishes habit; paid personal plans buy capacity and capability; Business and Enterprise add governance, company context and controls. The ladder maps well to increasing stakes, although overlap between plans is growing. |
| Pricing Metric | C+ | Subscriptions still bundle activities with very different compute costs and value. Higher limits control supplier cost, but a rate limit is not a customer-visible value meter. OpenAI's newer credit system points toward a better answer. |
| Operationalization | B+ | OpenAI already supports shared credits for supported agentic features, auto top-up and spend controls, while Business adds budgeting and administration. The weakness is fragmentation: credits are not yet the universal language for autonomous work. |
The scorecard shows what OpenAI gets right: packaging increasingly separates ordinary access from expensive work. Free can be generous because it is not being asked to carry unlimited frontier compute. Business can justify a seat component because a business workspace, identity controls and administration genuinely attach to users.
What OpenAI gets wrong is more subtle. The subscription still absorbs too many different economics.
Consider two Plus customers. One uses ChatGPT for short writing questions throughout the week. Another runs deep research, coding agents, file analysis and long reasoning jobs. Their monthly subscription can be identical while the cost OpenAI incurs and the value each customer receives are radically different. OpenAI manages that problem with allowances and rate limits.
From a product-operations perspective, limits are sensible. From a pricing perspective, they are incomplete. Customers discover the economic boundary only when they hit it.
OpenAI's own credit architecture reveals where the company is heading. As of August 2026, OpenAI allows customers on supported personal plans to buy credits after included allowances for supported agentic usage, with a shared balance, optional automatic top-ups and spending limits. That mechanism should become the centre of the next pricing reset rather than remain an extension attached to selected workloads.
Agentic work is now too autonomous for a flat subscription alone
The Agentic Monetization Spectrum, or AMS, helps determine how far an AI product should move away from classic per-user pricing. It examines three dimensions. Zero-human ability asks how much work the system can complete without continuous human effort. Operational domain measures whether it performs one narrow task, owns a complete workflow or works across several functions. Output/cost ratio asks how quickly the customer value of additional output rises relative to the compute cost required to produce it. As autonomy, breadth and the value-to-cost gap rise, a human seat becomes a weaker anchor and usage or outcome pricing becomes stronger.
ChatGPT's agentic layer now scores very differently from the chatbot that launched in 2022.
| AMS dimension | Score | Why it matters for ChatGPT |
|---|---|---|
| Zero-human ability | 2 / 3 | Users still initiate and review work, but research and coding systems can execute multi-step tasks between those points. |
| Operational domain | 3 / 3 | The same commercial product spans research, coding, analysis, documents, connected company tools and workflow execution. |
| Output/cost ratio | 2 / 3 | The value of a completed research or coding task can greatly exceed inference cost, but the ratio varies too much by task to support a universal outcome price. |
| Overall | 7 / 9 | ChatGPT has moved far enough towards autonomous work that usage should become the primary incremental meter, but not far enough for one general outcome fee. |
A 7/9 AMS position explains why neither extreme works. Pure seat pricing leaves too much variation inside a flat fee. Pure outcome pricing would be premature.
HubSpot can define a resolved support conversation. Salesforce can count an agent action. Those units have observable endpoints. A ChatGPT research task might save ten minutes or ten days. A Codex run might fix a trivial bug or contribute to a commercially valuable system. Attempting to calculate a percentage of the user's outcome would create arguments rather than alignment.
Credits fit the middle. They allow price to scale with machine work without pretending OpenAI can measure the customer's economic outcome.
The next reset should put one shared credit meter behind autonomous work
Monetizely's position is therefore specific: OpenAI should keep making baseline ChatGPT better for Free, retain subscription prices for access and predictable included capacity, and make a single shared credit meter for autonomous work the primary incremental paid metric.
Not tokens. Customers do not buy tokens.
Not time. A slow agent should not earn more revenue because it is inefficient.
Not arbitrary feature unlocks. The market will keep turning yesterday's premium AI features into tomorrow's table stakes.
The customer-facing unit should be a shared credit balance used when ChatGPT leaves ordinary conversation and starts doing substantial work on the customer's behalf. OpenAI has already built much of the operational foundation for this approach.
The commercial architecture could remain simple:
| Package | What the fixed price buys | Primary meter once included autonomous work is used |
|---|---|---|
| Free | Strong everyday ChatGPT, search and selected tools, with tighter advanced-use allowances; advertising can monetise eligible free usage. | Small included credit allowance; upgrade when meaningful autonomous usage begins. |
| Personal paid plans | Better models, priority, higher limits, ad-free use and a larger included allowance. | Shared credits for autonomous work. |
| Business | Seat-based workspace, administration, security, company context and pooled included capacity. | Workspace credit pool, with budget controls and automatic top-up. |
| Enterprise | Contractual controls, larger context, security, support and negotiated commitments. | Committed credit pool with volume rates and the same underlying meter. |
The synthesis is important: the seat remains useful for the workspace, but credits become the primary meter for additional machine labour.
OpenAI should also make the meter legible. Before a long-running job starts, ChatGPT can tell the user the expected credit range. A business administrator should be able to set limits by workspace or workload. Research, coding and other agents should draw from the same wallet rather than teaching customers a new currency for every product.
Such a reset would solve the apparent contradiction at the centre of OpenAI's strategy. Free can keep improving without destroying monetisation because the company would no longer depend on feature deprivation to create willingness to pay. The better Free becomes, the more people discover workflows worth delegating. The more work they delegate, the more naturally paid usage grows.
Operators should stop treating AI features as durable paywalls
OpenAI's strategy carries a wider warning for B2B software. Generative AI has shortened the commercial life of feature differentiation. A capability that supports a premium SKU this year can be bundled by a platform vendor next year and expected for free the year after.
The answer is not to protect the feature more aggressively. It is to move pricing closer to the durable source of value.
For operators making that transition, we would make five decisions now:
Model the business on the assumption that baseline AI intelligence approaches zero incremental price. Decide which revenue streams survive when good text generation, search, analysis and multimodal input are expected in every product.
Distinguish assistance from delegated work in product telemetry. Measure when users stop asking the software for advice and start asking it to complete multi-step work. That boundary is where a new pricing meter becomes economically defensible.
Build one consumption language before autonomous products proliferate. Customers will tolerate credits more readily than five unrelated units for research, coding, prospecting, analysis and workflow execution.
Let the free product become genuinely useful rather than deliberately frustrating. Upgrade pressure should come from users wanting more successful work completed, not from withholding capabilities that competitors will soon make standard.
Keep attention monetisation separate from work monetisation. Advertising can subsidise broad consumer access; business buyers should pay predictably for capacity, control and autonomous execution. Mixing the two weakens the trust required for higher-value enterprise use.
OpenAI's bet is ultimately larger than freemium. It is betting that intelligence itself will commoditise faster than the demand for intelligence can be exhausted. If that is correct, charging people merely to encounter the technology is the smaller opportunity.
The bigger opportunity begins when users hand the technology a job.
Assumptions
Current plan features and list prices are stated as published on primary vendor sources available on 21 August 2026; regional prices, negotiated enterprise discounts, temporary promotions and taxes are excluded unless noted. AMS scores and 5-Step Framework grades are Monetizely judgements rather than vendor-reported measures. OpenAI is privately held and does not publish a public-company 10-K or a regular public quarterly earnings-call transcript, so none has been invented or substituted; historical OpenAI pricing is anchored to dated Wayback captures plus OpenAI's own pricing, product and Help Centre records.
Footnotes
Monetizing Agentic AI: https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
OpenAI, archived ChatGPT Plus announcement, 1 February 2023: https://web.archive.org/web/20230201181831/https://openai.com/blog/chatgpt-plus/
OpenAI, archived ChatGPT pricing page, 10 January 2024: https://web.archive.org/web/20240110204456/https://openai.com/chatgpt/pricing
OpenAI, ChatGPT Plus announcement: https://openai.com/index/chatgpt-plus/
OpenAI, Introducing ChatGPT Team, 10 January 2024: https://openai.com/index/introducing-chatgpt-team/
OpenAI, GPT-4o and more tools to ChatGPT Free, 13 May 2024: https://openai.com/index/gpt-4o-and-more-tools-to-chatgpt-free/
OpenAI, Introducing ChatGPT Search, 31 October 2024, with subsequent availability updates: https://openai.com/index/introducing-chatgpt-search/
OpenAI, Introducing Deep Research, February 2025, with 24 April 2025 usage update: https://openai.com/index/introducing-deep-research/
OpenAI, ChatGPT Free Tier FAQ, current 21 August 2026: https://help.openai.com/en/articles/9275245-using-chatgpt-s-free-tier-faq
OpenAI, GPT-5.6 in ChatGPT, current 21 August 2026: https://help.openai.com/en/articles/20001354
OpenAI, Introducing ChatGPT Go, 16 January 2026: https://openai.com/index/introducing-chatgpt-go/
OpenAI, Ads in ChatGPT, current 21 August 2026: https://help.openai.com/en/articles/20001047-ads-in-chatgpt
OpenAI, Business and Enterprise pricing, current 21 August 2026: https://openai.com/business/pricing/
OpenAI, Using Credits for Flexible Usage in ChatGPT, current 21 August 2026: https://help.openai.com/en/articles/12642688-using-credits-for-flexible-usage-in-chatgpt-freegopluspro-sora
Google Workspace, The future of AI-powered work for every business, 15 January 2025: https://workspace.google.com/blog/product-announcements/empowering-businesses-with-AI
Google Workspace pricing, current 2026: https://workspace.google.com/pricing
Microsoft, Enabling agents in Microsoft 365 Copilot Chat, January 2025: https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/enabling-agents-in-microsoft-365-copilot-chat/
Microsoft, Microsoft 365 Copilot pricing announcement, 18 July 2023: https://www.microsoft.com/en-us/microsoft-365/blog/2023/07/18/introducing-bing-chat-enterprise-microsoft-365-copilot-pricing-and-microsoft-sales-copilot/
Microsoft, FY2026 Q3 earnings call, 29 April 2026: https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q3
Salesforce, Agentforce pricing, current 2026: https://www.salesforce.com/in/agentforce/pricing/
Salesforce Investor Relations, Agentforce 2.0 announcement, December 2024: https://investor.salesforce.com/news/news-details/2024/Introducing-Agentforce-2.0-The-Digital-Labor-Platform-for-Building-a-Limitless-Workforce/default.aspx
HubSpot, Customer Platform pricing, current 2026: https://www.hubspot.com/pricing/suite
HubSpot Investor Relations, HubSpot Credits, 8 May 2025: https://ir.hubspot.com/news-releases/news-release-details/hubspot-credits
HubSpot, Customer Agent and Prospecting Agent outcome pricing, 13 April 2026: https://www.hubspot.com/company-news/hubspots-customer-agent-and-prospecting-agent-now-you-pay-when-the-task-is-complete