LLM API Pricing Diverges as the Era of Subsidized Low Prices Comes to an End

    LLM API prices fell sharply over the past year, but in 2026 some providers began raising monthly fees, cutting free quotas, and discontinuing entry-level plans. This article explains the three-tier monthly plan structure, compares domestic and overseas pricing, examines recent price increases and promotion changes, and offers practical advice for developers choosing and managing API plans.

    Over the past year, LLM API providers competed fiercely on price. Prices dropped from bargain-basement levels to free credit giveaways, as vendors spent heavily to grow their developer ecosystems. In 2026, the market is shifting. Some platforms have raised monthly fees, free quotas have shrunk substantially, and some products have stopped selling their entry-level tiers. These signs suggest LLM API services are moving from “burning money for growth” to “refined operations.”

    1. Clear Tiers in Pricing

    Monthly plans from mainstream LLM API services can be roughly divided into three tiers:

    Entry tier (RMB 30 to 50 per month): Aimed at individual developers, learners, and light application use. Typical features include basic model capability, limited context length, and relatively low call-volume quotas. For example, one provider’s RMB 29 per month plan focuses on the lowest-barrier trial experience. Another major platform’s RMB 39 per month plan also bundles high-frequency toolchains such as agent orchestration, PPT generation, and code assistance, which makes it look more cost-effective.

    Professional tier (about RMB 150 per month): Aimed at small teams and independent developers with meaningful workloads. One platform’s RMB 149 per month plan emphasizes engineering capability, including longer context support, higher API call quotas, and higher-priority inference scheduling. At this price point, products mainly differ in whether they include a dedicated agent runtime environment, private knowledge base integration, and support for more complex multi-turn conversation workflows.

    Enterprise tier (hundreds to thousands of RMB per month): This tier falls outside the low-price scope of this article. It usually involves private deployment or dedicated resource pools, with prices set through sales negotiation. It is not covered here.

    The boundary between the entry and professional tiers is blurring. Some providers add more usage to entry plans at the same price to make them more attractive. Others deliberately restrict key capabilities in entry plans, such as disabling web search or limiting file upload size, to push users toward upgrading.

    2. Why Domestic and Overseas Prices Differ

    Comparing the same provider’s domestic and overseas prices reveals a notable pattern. Not every product costs more overseas. Some cost less abroad.

    One pattern is “lower in China, higher overseas.” For example, a leading provider’s lowest domestic monthly fee is RMB 49 (about USD 7), while its overseas price is USD 18. The gap exceeds 2.5 times. The opposite also occurs. One provider’s overseas entry Plus plan costs USD 9.99 (about RMB 72), while the comparable domestic plan costs RMB 99. The overseas price is about 27% lower.

    This inversion does not reflect generosity. Several factors drive it.

    First, currency conversion is not a simple calculation at the real-time exchange rate. Providers typically price based on purchasing power parity in the target market, competitors’ price ranges, and what local developers are willing to pay. The North American market has more mature payment habits for SaaS and API services, but competition there is also fiercer. Some providers therefore use low prices to attract early users.

    Second, quota design is a hidden lever. Two plans with nearly identical prices can differ greatly in daily call limits, token throughput caps, concurrent connections, and context cache size. One developer tested a domestic RMB 29 plan and found its quota was only about 40% of an overseas USD 10 plan. Calculated per token, the overseas plan was the better deal.

    Third, subsidy strategies differ by stage. After the intense price war of 2024 and 2025, some domestic providers reduced subsidies and returned to more rational pricing. Overseas markets are still in a market-share land-grab phase, and their low-price strategies have not yet reached a turning point.

    3. Price Increases and Discontinuations as Industry Signals

    Several signals in the past month deserve attention.

    Price increases generally exceed 30%. One well-known platform raised its Pro plan from about RMB 114 to RMB 149 per month, an increase of more than 30%. Another provider stopped selling its cheapest Lite plan, raising its lowest entry point from the teens of RMB to the RMB 40 tier.

    Free quotas and limited-time flash sales have become the norm. Promotions that once gave away millions of tokens have largely disappeared. In their place are limited offers that users must grab at a specific time each day. This is essentially a scarcity-driven sales tactic. It helps keep prices stable and filters for users with real, high-frequency demand.

    First-purchase offers are now limited. Most platforms previously offered new users deals such as “first month for RMB 1” or “50% bonus on first top-up.” Those policies have now been canceled or replaced with daily caps and first-come, first-served access, which makes them much harder to get.

    The core logic is that LLM APIs are shifting from traffic gateways to profit centers. Once market structure stabilizes, providers will inevitably shift their focus from user growth to monetization. For developers, this means the assumption formed during the low-price period, that APIs can be called freely, needs to be corrected soon.

    4. Pitfalls Developers Should Avoid

    Here are some lessons for this period of pricing adjustment.

    First, check the API key type and permission scope carefully. Some platforms distinguish between “dedicated keys” and “general keys.” Using the wrong one causes charges to be billed per usage instead of drawing on the plan’s included quota. For example, one cloud provider requires keys prefixed with sk-sp- to use plan resources. Using a regular key may cost more than five times as much. Read the relevant API documentation during setup, and explicitly set environment variables in code to avoid production incidents.

    Second, focus on quota consumption rate, not just price. Two plans with the same monthly fee may differ: one supports 1 million tokens per day, while the other supports only 400,000. In high-volume call scenarios, the actual cost of the second plan may be as much as 1.5 times higher. Before choosing a plan, simulate costs using your application’s historical call data rather than being misled by a low headline price.

    Third, plan migration early if you rely on a plan being discontinued. If your existing workload depends on an entry plan that is about to be discontinued, start planning the migration now. Existing subscribers may get a grace period, but new users cannot buy the plan, and renewal rules may change. Back up conversation logs, fine-tuning parameters, and evaluation benchmarks in advance to lower switching costs.

    Fourth, prioritize flexible plans without mandatory lock-in. Some plans require monthly subscriptions that cannot be downgraded mid-term, while others allow pay-as-you-go and subscription billing to coexist. For workloads with significant demand fluctuation, a combination of a subscription baseline and pay-as-you-go overflow is safer.

    5. Selection Guidelines: Avoid Overpaying and Avoid Speculative Bets

    Based on current market conditions, these selection principles have broad relevance.

    • Monthly budget under RMB 50: Prioritize plans that include high-frequency tools such as web search and code execution, rather than focusing only on the model’s capability ceiling. For most personal projects, the value of the toolchain at the RMB 30 to 40 tier far outweighs the difference in model parameter counts.

    • Monthly budget of RMB 100 to 200: Check whether the plan includes a dedicated agent runtime environment, private knowledge base hosting, and multi-user team collaboration. Differences in engineering capability mainly appear in these surrounding services rather than in basic conversation quality.

    • Highly cost-sensitive: Consider running non-core workloads on overseas plans to benefit from exchange rate and price differences. Check data compliance requirements, and confirm that the business scenario allows cross-border data transfer.

    LLM API pricing is moving from “traffic thinking” to “value thinking.” Rather than complaining that free perks are harder to get, enterprise developers should establish refined cost management now. Include API calls in the application’s resource monitoring, and manage token consumption the way you manage database connections and storage. This is the most reliable response to price increases. The free lunch is over, but precise budget allocation will always matter more than chasing the lowest price.