When evaluating LLM price cuts, decision-makers should not start by asking whether user counts increased. A better first question is whether the unit economics of their own workloads have changed.

Based on the public evidence cited here, it is not confirmed that OpenAI had 1 billion active users and 2 million enterprise customers after the price cut. The verifiable official statements are different: a projection that it would “soon reach 1 billion weekly active users,” “more than 1 million business customers,” and “more than 2 million weekly Codex users.” For that reason, this issue is better read not as confirmation of a user-growth milestone, but as a signal to examine how lower-cost inference could affect product, procurement, and infrastructure decisions.

What matters more than the numbers is the direction of the price cuts

It is hard to explain LLM inference price cuts as only a short-term discount campaign. The research in the reference materials suggests that major providers are prioritizing broader adoption and often lower prices after competitive launches. This resembles an early software-market pattern: reduce unit prices to increase usage, then use that usage to become more deeply embedded in product habits, developer ecosystems, and enterprise workflows.

However, the cycle of “price cuts → user surge → revenue growth → investment in next-generation models” should be treated as a possible strategy, not as a causal relationship proven by the evidence cited here. Without disclosure of key variables such as the active-user measurement period, duplicate-account handling, paid conversion rates, and average enterprise customer spend, it is difficult to determine whether demand growth has improved profitability.

For practitioners, the more useful question is not “Prices have fallen, so can we use more?” It is. “Has the price decline lowered the marginal cost of previously blocked use cases enough to make them viable?”

The comparison axis among closed APIs, open source, and self-hosting is changing

Price cuts do not only affect competition among closed LLM providers. They also require teams to recalculate the case for open-source models and self-hosting.

First, among closed APIs, pressure will increase to expand low-cost models and differentiate by use case. If a cheaper model can handle the same task well enough, high-performance models are less likely to remain the default for every request. They become more suitable for exception handling, difficult reasoning, and quality review. Product teams should consider routing requests by difficulty instead of choosing one “best model” for all cases.

Second, the case for open-source models becomes more specific rather than weaker. According to the research in the reference materials, open-source models may already be cheaper than closed models with comparable intelligence. But if closed API prices fall further, open-source adoption becomes harder to justify on token price alone. The stronger reasons are likely to be data control, customization, latency, domain-specific optimization, and avoiding provider lock-in.

Third, self-hosting requires especially careful calculation. In a case study of enterprise coding agents, a 99.3% prompt-caching hit rate reduced realized API cost by 88.6% to $0.57 per 1 million tokens, below the amortized unit cost of $2.83 for a shared on-premises slice. This does not mean the same result applies to every workload. It does suggest a practical standard: do not assume self-hosting is cheaper until API costs have been recalculated after caching, batching, and low-cost model routing.

Decision rule: compare the “realized unit cost,” not the average unit price

The decision rule enterprises can use in this phase of price cuts is as follows.

First, divide current usage into three categories: tasks with many repeated prompts and large caching benefits, tasks with low quality sensitivity that low-cost models can handle sufficiently, and tasks where failure costs are high and high-performance models are required. This distinction disappears if teams look only at listed input and output token prices. Actual cost should include caching, batch discounts, model routing, retry rates, and post-processing costs.

Next, filter self-hosting candidates based on sustained utilization. Owning GPUs or servers is not automatically a cost advantage. If utilization is low or traffic is volatile, fixed costs raise the unit cost. Conversely, if usage is stable, data-control requirements are strong, and operational capabilities exist, self-hosting may still be worth considering.

Finally, provider selection should not be based only on “the cheapest model today.” It should also account for whether price declines can be reflected in the product architecture. In a market where prices continue to change, an architecture that makes model replacement and routing easier is more likely to preserve cost savings than one fixed to a single model.

What to believe and what to reserve judgment on in this announcement

What can be confirmed from official materials is that OpenAI is emphasizing a large user base and enterprise usage. The cited materials also present figures such as more than 1 million business customers and more than 2 million weekly Codex users. By contrast, “reaching 1 billion active users” and “2 million enterprise customers” are difficult to treat as confirmed based only on the verification results provided here.

Even with those limits, the price cut itself is a meaningful signal. As LLM costs fall, AI features can move from “expensive features for some users” toward broader embedding in core workflows. But the benefits are not automatic. Teams need to identify cacheable tasks, route work by model difficulty, and compare self-hosting fixed costs against realized API unit costs before treating price cuts as a product advantage.

Further Reading

• Operational patterns in LLM fraud defense
• UrbanDS and the data discovery problem in LLM agents
• Penelope’s tradeoff between latency and auditable reasoning
• Deployment criteria for nuclear LLMs
• Why Legal Structure Chunking Matters for EU AI Act

References

• OpenAI raises $122 billion to accelerate the next phase of AI - openai.com
• The state of enterprise AI | OpenAI - openai.com
• Menu Pricing of Large Language Models - cowles.yale.edu
• Inference Economics of Enterprise Coding Agents: A Case Study of Cloud vs. On-Premise LLMs - arxiv.org