OpenAI said on Aug. 7 that ChatGPT had reached 1 billion weekly active users, as it opened unlimited text chats with its GPT-5.6 Luna model to free users. The move places a lower-cost model at the center of ChatGPT’s mass-market offering while reserving advanced reasoning and complex coding work for the company’s higher-end SOL-series models.
The announcement arrived as DeepSeek API users were told that broad price increases were approaching, underscoring a growing divide in how major artificial intelligence providers are handling the cost of serving rapidly expanding demand. DeepSeek had already introduced peak and off-peak rates in mid-July, with peak prices doubling, before issuing its Aug. 6 notification without a timetable or detailed rate card.
OpenAI’s decision gives free users unrestricted access to a model designed for routine conversations, basic writing, and standard translation. That could help the company maintain engagement across a user base that now exceeds the population of most online platforms, while directing users who need multi-step code analysis or heavier reasoning workloads toward paid products.
Free access supports scale, while paid products target heavier workloads
ChatGPT’s 1 billion weekly active users include 50 million paying subscribers, according to OpenAI’s figures. That equates to a paid conversion rate of roughly 5%, leaving the company dependent on a large free audience that creates substantial inference demand: the computing work required every time a model generates a response.
The Luna release suggests OpenAI is trying to make that economics more manageable by matching common tasks to a mid-tier model. Everyday chat and simple writing requests generally require fewer computational resources than software debugging, long-chain reasoning, or autonomous agent tasks that execute multiple steps.
OpenAI reported first-quarter revenue of $5.7 billion in 2026, alongside a non-GAAP operating loss margin of negative 122%. In practical terms, the company spent about $1.22 for every dollar of revenue during the period. It projected a full-year net loss of $14 billion.
Those figures show why user growth alone does not resolve the commercial challenge facing large language model providers. A free service can build distribution and feed demand for premium tools, but each additional active user also adds recurring serving costs. The pressure becomes sharper when users move from short text requests to long documents, code repositories, or agent-style workflows that can trigger many model calls.
OpenAI’s business is not solely tied to consumer subscriptions. Enterprise customers contributed more than 40% of revenue, according to the company’s reported figures, while enterprise API usage continued to grow. Codex, its coding product, reached 5 million weekly users. Its advertising product also reached an annualized revenue run rate of $100 million after six weeks, based on figures cited in the supplied material.
Together, those businesses offer paths to monetization beyond individual subscriptions, though they also depend on OpenAI being able to provide reliable capacity for commercial users.
DeepSeek demand has strained API performance
DeepSeek’s pricing warning follows a period of heavy reported usage for its V4 Flash model. V4 Flash ranked first on OpenRouter’s global weekly usage leaderboard and processed 7.22 trillion tokens over one week, according to the figures in the supplied material. A token is a unit of text processed by an AI model; long prompts, generated answers, and repeated agent actions all raise token consumption.
OpenCode data cited in the material showed V4 Flash processing 8 trillion tokens on Aug. 1 alone. Users also reported timeouts and slower performance during weekday peak periods, pointing to a mismatch between demand and the available computing capacity allocated to the service.
DeepSeek’s peak and off-peak pricing system is a direct response to that constraint. Higher rates during busy periods can encourage some customers to move non-urgent workloads to quieter hours, reducing pressure on servers without requiring the provider to deny access outright. A broader increase would raise the cost of API-based applications that rely on DeepSeek for customer support, text generation, coding assistance, or other automated functions.
The pricing changes also challenge the assumption that low-cost model access will remain stable as usage rises. Providers can lower prices through model efficiency and competition, but the infrastructure behind large-scale inference remains expensive: chips, networking equipment, electricity, cooling, data-center space, and engineering capacity all sit behind each API request.
Chinese consumer AI services test paid tiers
China’s Doubao has adopted a similar segmentation strategy from a consumer angle. The service launched paid plans on June 24 priced at 68 yuan, 200 yuan, and 500 yuan a month, while maintaining free access to basic functions.
Doubao was reported to have 345 million monthly active users. The supplied article said its average daily inference costs ran into tens of millions of yuan, while e-commerce commissions had not been sufficient to offset those expenses. That creates a familiar calculation for consumer AI platforms: free access expands reach, but premium tiers must capture enough high-value usage to cover the disproportionately high cost of power users.
The emerging pricing structure across providers has three layers. Basic question-and-answer services remain free or inexpensive to preserve broad access. Professional tools for tasks such as coding and data analysis are increasingly billed through subscriptions, usage limits, or API rates. Agent products, which can carry out longer workflows with many model calls, may eventually move toward outcome-based pricing, where customers pay for a completed task rather than raw token volume.
That progression reflects the different economics of AI workloads. A short request to summarize an email may need only a small amount of compute. An agent asked to analyze a codebase, search documentation, test fixes, and produce a report can consume far more memory, processing time, and model output.
OpenAI’s free Luna rollout and DeepSeek’s impending API increase therefore represent two sides of the same market response. One provider is expanding broad access through model tiering; the other is raising prices after demand appears to have pushed against capacity. The next phase of competition will depend less on headline user counts alone and more on which providers can keep everyday AI affordable while charging enough for intensive workloads to fund the infrastructure behind them.
Explore AI-powered trading automation with Toobit’s AI copy trading to mirror top strategies while focusing on high-value reasoning tasks.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
