Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
OpenAI made its cheapest ChatGPT model both faster and 80 percent cheaper this month, then gave it away for free. GPT-5.6 Luna, the budget tier of the model family OpenAI shipped on July 9, 2026, now costs $0.20 per million input tokens and $1.20 per million output tokens through the API, down from $1.00 and $6.00. Free and Go-tier ChatGPT users got Luna as their new default model starting the week of August 6, with unlimited text chats and the “Think” reasoning toggle following through August 10, according to OpenAI’s own release notes and TechCrunch’s reporting.
The timing is not a coincidence. OpenAI has watched its share of the AI chatbot market erode for more than a year. Similarweb pegged ChatGPT’s global web traffic share at 53.9 percent in May 2026, down from roughly 79 percent twelve months earlier. Sensor Tower’s app data tells a similar story: ChatGPT’s slice of AI assistant app users fell below 50 percent in March 2026 and sat at 46.4 percent by May, even as Google’s Gemini climbed to 27.7 percent, according to TechCrunch’s coverage of the Sensor Tower figures. ChatGPT still crossed 1 billion monthly active users in June 2026, the fastest any consumer app has reached that mark. A free, unlimited, meaningfully cheaper GPT-5.6 Luna reads like a direct answer to that squeeze.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
GPT-5.6 shipped as three separate models rather than one. Sol is the flagship, built for users who want maximum reasoning depth. Terra sits in the middle, tuned for everyday work like planning, writing, and research. Luna is the cost-efficient option, designed to handle high-volume, lower-complexity requests cheaply.
Until this month, free ChatGPT users were capped on how many messages they could send per day and ran on GPT-5.5 by default. That changes with Luna. OpenAI is now giving Free and Go-tier users unlimited text chats on Luna, plus access to a “Think” button that lets them ask the model to spend more reasoning effort on a hard question before it answers. Paying Plus and Pro subscribers keep their access to Sol, now retuned for speed on quick tasks, and gain a thinking slider that lets them dial reasoning effort up or down per message.
OpenAI says the retuned Sol is more reliable with facts and gives more focused answers, and that Luna produces 62 percent fewer factual errors than GPT-5.5-Instant, the model it effectively replaces for free users. Those numbers come from OpenAI’s own release notes rather than independent benchmarking, so they are worth treating as a starting point rather than a verdict.
The price cuts landed on July 30, a week and a half before the free-tier switch, and they were uneven by design. Sol, the flagship tier, held steady at $5.00 per million input tokens and $30.00 per million output tokens, the same price OpenAI set at launch. Terra dropped 20 percent, from $2.50/$15.00 to $2.00/$12.00 per million tokens. Luna took the deepest cut by far, an 80 percent reduction that took GPT-5.6 Luna from $1.00/$6.00 down to $0.20/$1.20 per million tokens.
OpenAI has attributed the cut to efficiency gains in how it routes requests across models and inference hardware, saying tighter integration between models, tools, and compute lets more traffic run on cheap infrastructure without forcing everyone onto the pricier Sol tier. The table below shows the full shift.
Judged by OpenAI’s own release cadence, this rollout moved fast. The GPT-5.6 family launched July 9. The Luna and Terra price cuts followed on July 30. The free-tier default switch to Luna landed August 6, with the unlimited-chat cap removal and Think button finishing their rollout by August 10. That is three material changes to the same model line inside four weeks.
Compare that to the cadence around GPT-5.6 Sol’s earlier rollout, when OpenAI initially gated access to roughly 20 partner firms before opening it to everyone, a staggered release that played out over months. Luna’s path looks different: ship, cut price, make it the default, in rapid succession. That sequencing suggests OpenAI now treats its free tier less as a cost to tolerate and more as a retention lever to pull hard and quickly once competitive pressure builds.
Free users do not generate direct revenue, so subsidizing them this heavily is a deliberate bet, not an accident. Three forces appear to be driving it.
First, distribution. ChatGPT’s app crossed 1 billion monthly active users in June 2026, a scale advantage OpenAI has no interest in shrinking. Every free user who leaves for Gemini or Claude is a smaller top of funnel for eventual Plus and Pro conversions.
Second, unit economics have genuinely improved. An 80 percent price cut on Luna only makes sense if OpenAI’s actual cost to serve each token dropped by a comparable amount, whether through better routing, cheaper inference hardware, or model compression. OpenAI has pointed to infrastructure efficiency rather than a temporary promotional discount, which suggests the company expects the lower price to hold.
Third, rivals are no longer standing still. Google’s Gemini 3.6 Flash and Gemini 3.1 Pro give away a capable free tier bundled with Search and Workspace. Anthropic’s Claude, historically the priciest of the major labs, has been cutting its own prices too. Giving away a materially better free product is one of the few levers OpenAI can pull that rivals cannot immediately copy without matching the underlying infrastructure investment.
The numbers explain the urgency. Similarweb’s web-traffic data shows ChatGPT holding 53.9 percent of global AI chatbot traffic in May 2026, down from about 79 percent a year before, a 25-point drop in twelve months. In the United States specifically, ChatGPT holds 58.3 percent of web visits, with Gemini at 19.3 percent and Claude at 13.4 percent, on the same tracking.
App-usage data from Sensor Tower shows the same trend from a different angle. ChatGPT’s share of AI assistant app users fell below 50 percent for the first time in March 2026 and had slipped to 46.4 percent by May. Gemini, over the same period, rose to 27.7 percent of app share, up from a low-teens share a year earlier, largely on the back of its integration into Android and Google Search.
None of this means ChatGPT is shrinking in absolute terms. The 1-billion-user milestone says otherwise. It means the market is growing faster than ChatGPT’s slice of it, and Google in particular is converting its existing distribution into AI assistant share at a pace OpenAI cannot match through product quality alone. A free, unlimited, meaningfully improved Luna is a direct lever against that trend.
Price alone does not decide which model wins a given task, but it decides which model gets tried first for high-volume, cost-sensitive work. Here is how GPT-5.6’s three tiers line up against the current price lists at Anthropic, Google, and DeepSeek, based on each company’s published API rates as of early August 2026.
At $0.20 per million input tokens, GPT-5.6 Luna undercuts every mainstream proprietary model on this list except DeepSeek’s V4-Flash, which OpenAI is unlikely to match given DeepSeek’s much lower reported training and serving costs. Luna prices below Gemini 3.6 Flash ($1.50/$7.50) by a wide margin and sits nowhere near Claude Sonnet 5’s introductory $2.00/$10.00 rate. For a developer running high-volume, latency-tolerant workloads such as support-ticket triage, first-pass content generation, or simple classification, Luna is now cheap enough to make the build-versus-buy math against open-weight models look far less obvious than it did a month ago.
Cheap does not mean equivalent. Luna is explicitly the budget tier of the GPT-5.6 family, positioned below Terra and well below Sol on reasoning depth. OpenAI’s own materials frame Sol, not Luna, as the model for complex, multi-step work. Claude Opus 4.8 and Gemini 3.1 Pro remain priced and positioned as premium reasoning models, and neither Anthropic nor Google has signaled plans to give away their top-tier models the way OpenAI now gives away Luna. The practical read: Luna competes hardest against other labs’ cheap, high-volume tiers, not against their flagship reasoning models. Anyone routing complex agentic or coding work still has reason to reach for Sol, Opus, or Gemini 3.1 Pro instead.
OpenAI’s language says free and Go-tier users get unlimited text chats on Luna, but the rollout is staged and the fine print matters. The default-model swap to Luna landed first, in the week of August 6. Unlimited chats and the Think button follow in the week of August 10, meaning the full experience was not available to every free user on day one.
“Unlimited” also applies specifically to text chat. It does not obviously extend to image generation, voice mode, or file uploads, none of which OpenAI has said will lose their existing usage caps for free accounts. Heavy users who lean on multimodal features will likely still hit limits that text-only users no longer face.
There is also a practical question any free product like this eventually confronts: what happens when unlimited use meets real infrastructure cost. OpenAI has not published a fair-use policy alongside the unlimited claim, and companies offering “unlimited” tiers have historically added soft caps once usage patterns settle. Whether that happens here is one of the more interesting open questions in this rollout.
For developers and enterprise buyers, the more consequential number may be the API pricing, not the ChatGPT free tier. At $0.20/$1.20 per million tokens, Luna becomes viable for workloads that were previously routed to smaller open-weight models purely on cost grounds, things like high-volume customer support triage, log summarization, or first-pass content moderation.
Terra’s smaller 20 percent cut, to $2.00/$12.00, positions it as the default choice for teams that want GPT-5.6-family quality without Sol’s premium price. Sol itself, unchanged at $5.00/$30.00, remains the tier OpenAI is pitching for agentic and coding workloads that need maximum reliability.
Switching a production workload to Luna is a one-line change in most existing OpenAI SDK integrations:
For teams already on GPT-5.5, the migration mainly requires re-testing prompts against Luna’s behavior rather than any structural code change, since the API surface stays consistent across the GPT-5.6 family.
ChatGPT’s free tier has changed models several times since the product launched in November 2022 on GPT-3.5. GPT-4o became the default in May 2024, giving free users a genuinely capable multimodal model, if a usage-capped one. When GPT-5 arrived on August 7, 2025, OpenAI initially pulled GPT-4o from ChatGPT entirely. Users who preferred the older model pushed back hard enough that OpenAI reinstated GPT-4o access for paying subscribers within days.
That episode looks relevant now. It showed OpenAI that pulling a familiar free model without warning carries real reputational cost, which may help explain why the Luna rollout leaned so heavily on public communication: detailed release notes, an OpenAI blog post, and direct outreach to press, instead of a quiet swap. A year after the GPT-5 backlash, OpenAI is treating a free-tier model change as a headline event rather than a routine update, itself a signal of how much weight the company now puts on free-user retention.
Canadian ChatGPT users get the same Luna rollout as everyone else, since OpenAI has not signaled any region-specific gating for this change. For Canadian developers billing in USD through the API, Luna’s new $0.20/$1.20 per million token rate converts to roughly CA$0.28/CA$1.67 at the USD/CAD rate of about 1.39 in early August 2026, still a steep drop from Luna’s pre-cut cost.
The bigger question for Canada is competitive. Claude adoption has grown sharply in the Canadian market this year, and ChatGPT’s own Canadian market share has been under pressure even as Canadian enterprise AI spending keeps climbing. A free, unlimited, meaningfully improved ChatGPT tier raises the bar Claude and Gemini need to clear to win over Canadian consumers who have not yet picked a primary AI assistant, even as it does little to change the calculus for enterprise buyers who select models on data residency, security, and procurement terms rather than consumer pricing.
A few things about this rollout remain unresolved. OpenAI has not said how long the current Luna pricing will hold, and the company has cut and adjusted GPT-5.6 prices on short notice twice already. Free-tier “unlimited” claims also tend to attract abuse: automated scraping, bulk content generation, and API-key resale schemes dressed up as personal use. OpenAI has not detailed what enforcement looks like against that kind of misuse.
There is also a quality-control risk. Pushing a cheaper, faster model to over a billion monthly users raises the stakes on Luna’s error rate. OpenAI’s claimed 62 percent reduction in factual errors versus GPT-5.5-Instant is a comparison against a specific model OpenAI is retiring from that role, not an independent, third-party benchmark against current rivals. Until outside evaluators publish their own numbers, that figure should be read as OpenAI’s internal claim rather than settled fact. Independent benchmarks typically arrive within a few weeks of a wide model release, so a clearer picture should emerge before September 2026.
GPT-5.6 Luna is the budget-tier model in OpenAI’s GPT-5.6 family, released July 9, 2026 alongside the flagship Sol and mid-tier Terra. As of August 2026, it is the default model for Free and Go-tier ChatGPT users.
$0.20 per million input tokens and $1.20 per million output tokens, after an 80 percent price cut on July 30, 2026. It previously cost $1.00/$6.00 per million tokens.
OpenAI says free and Go-tier users get unlimited text chats on Luna, rolling out through the week of August 10, 2026. The unlimited claim applies to text chat specifically. OpenAI has not detailed limits for image generation or voice mode.
Sol is the flagship model for complex reasoning, priced at $5.00/$30.00 per million tokens. Terra is the balanced mid-tier at $2.00/$12.00. Luna is the cheapest, fastest tier at $0.20/$1.20, built for high-volume, lower-complexity tasks.
Yes. Luna is now the default model for Free and Go-tier ChatGPT users, replacing GPT-5.5 in that role.
Luna, at $0.20/$1.20 per million tokens, undercuts Google’s Gemini 3.6 Flash ($1.50/$7.50) and Anthropic’s Claude Sonnet 5 (from $2.00/$10.00) by a wide margin. Only DeepSeek’s open-weight V4-Flash, at $0.14/$0.28, prices lower.
Yes. OpenAI has not announced any region-specific restrictions on Luna’s API or ChatGPT rollout for Canada.
OpenAI has not committed to a fixed timeline for the current pricing. The company has adjusted GPT-5.6 pricing twice within a month of launch already, so further changes are plausible.
Marcus Chen is a senior editor at Tech Insider, where he leads coverage of the US online gaming market, including sweepstakes and social casinos, alongside consumer technology. He evaluates operators on their published terms, licensing and RNG certifications, stated redemption policies, and corroborating independent reporting, and writes plainly about what the evidence supports. Tech Insider does not run first-party money tests and does not gamble with reader funds. Marcus has reported on the technology and online-gaming industries for more than a decade.
Tech Insider delivers in-depth coverage of the technologies shaping the future: AI, cybersecurity, cloud computing, hardware, and the trends that matter.
