Douyin Launches Paid Subscription Tiers, Ending Free AI Model Era

On May 4, Douyin, ByteDance’s core large language model, announced the introduction of paid subscription plans alongside its free tier, according to reporting by澎湃新闻 (The Paper). The move marks a potential end to the free-access era for Chinese AI models and reflects growing demand for advanced capabilities in complex tasks.

Subscription Pricing and Tiers

Douyin will offer three subscription levels with monthly and annual billing options:

• Standard: ¥68 per month (¥688 annually)
• Enhanced: ¥200 per month (¥2,048 annually)
• Professional: ¥500 per month (¥5,088 annually)

The pricing structure was discovered on Douyin’s AppStore page, which included a service statement detailing the paid offerings.

Official Response and Testing Phase

Douyin’s official team responded to The Paper, stating: “Douyin will continue to provide free services. Building on free offerings, Douyin is exploring additional value-added services to meet differentiated user needs. Related plan details are currently in testing phase and will be released through official channels upon formal launch.”

Target Use Cases and Paid Feature Focus

According to sources close to Douyin, paid features will concentrate on complex tasks and productivity scenarios, including:

• PowerPoint generation
• Data analysis
• Video and film production

As model capabilities advance, Douyin can now handle increasingly complex, high-value tasks. However, these tasks require greater computational resources and inference time, prompting the company to introduce paid services to meet demand while maintaining free access for everyday user needs.

Market Context: Token Usage and Growth

As of March 2025, Douyin’s large language model reached 1.2 quadrillion tokens in daily usage, representing 1,000x growth since its May 2024 launch and doubling over the previous three months, according to Volcano Engine (ByteDance’s cloud division). On Volcano Engine’s platform, the number of enterprises with cumulative token usage exceeding one trillion has grown from 100 companies at year-end 2024 to 140 companies.

Volcano Engine VP Tan Dai previously addressed large model pricing strategy, noting that price increases reflect capability improvements: “Token price differences fundamentally reflect capability differences. Next-generation models are more powerful; per-token costs will rise, but the economic value created will increase proportionally. Price increases driven by improved model intelligence ultimately create greater value for customers.”

Competitive Landscape: Global and Domestic Precedent

Douyin’s paid model follows established patterns globally and domestically. OpenAI’s ChatGPT offers tiered access (Free, Go, Plus, Business, Enterprise) with regional pricing variations. Competitors Anthropic and Grok have also launched paid subscription versions.

In China, leading models Zhipu and DeepSeek already operate paid subscription models. However, pricing strategies diverge: Zhipu implemented three API price increases in 2025 alone. On February 12, Zhipu raised GLM Coding Plan pricing by 30% or more, citing “sustained strong market demand and rapid growth in user scale and call volume.” On March 16, Zhipu released GLM-5-Turbo with a 20% price increase, followed by a 10% increase on April 8 with GLM-5.1.

In contrast, DeepSeek announced significant price cuts on April 25, offering a limited-time 75% discount on V4-Pro API pricing. On April 26, DeepSeek further reduced cached input token prices across its entire API suite to one-tenth of original pricing, with Pro model discounts extending through May 5, 2026. Post-adjustment pricing: DeepSeek-V4-Flash at ¥0.02 per million cached input tokens and DeepSeek-V4-Pro at ¥0.025 per million tokens.

The divergent pricing strategies—Zhipu raising, DeepSeek cutting—reflect independent market judgments rather than an industry-wide trend.