Volcengine Price Drop: Get the Agent Plan for ¥9.9 — A Domestic Answer to Claude Max

Volcano Engine Drops AI Programming Subscription to ¥9.9/Month

On June 8, ByteDance’s Volcano Engine made the most straightforward move of the summer: both the entry-level and mid-tier tiers of the Agent Plan and Coding Plan were discounted to 25% of their original price. A ¥40/month plan is only ¥9.9/month for the first two months, while the ¥200/month tier drops to ¥49.9/month. The promotion runs from now until August 27 and applies to new purchases, upgrades, and renewals.

This is the first time among mainstream domestic AI vendors that an “Agent all-in-one” subscription has been priced close to the psychological threshold of Cursor’s free version. Considering Claude Max’s availability in China and exchange rate issues, the aggressiveness of the ¥9.9 pricing is self-evident—it’s not trying to lure Cursor users; it’s directly aiming to “stop you from topping up Claude.”

Four Tiers, Prices at a Glance

The promotion covers two product lines with four price tiers. Coding Plan focuses on pure programming scenarios; Agent Plan adds multimodal models and the Harness tool chain on top.

| Plan | Original Price | Promo Price | Discount | | --- | --- | --- | --- | | Coding Plan Lite | ¥40/month | ¥9.9/month | 25% | | Coding Plan Pro | ¥200/month | ¥49.9/month | 25% | | Agent Plan Small | ¥40/month | ¥9.9/month | 25% | | Agent Plan Medium | ¥200/month | ¥49.9/month | 25% |

From the third month onward, prices revert to the original rate, so strictly speaking, this is a “first two months trial price.” However, even at the regular ¥40, AI Coding subscriptions aren’t expensive—compare it to Claude Pro’s $20, where, according to official figures, domestic developers get several times the effective capacity of Claude Max, and at least five times for the Lite tier.

Notably, the promotion covers only the ¥40 and ¥200 tiers; the higher ¥500/month and ¥1,000/month tiers are excluded. Volcano’s intention is clear: attract new users and drive engagement, not subsidize heavy commercial use.

Agent Plan’s Trump Card: Model + Harness Dual Integration

Let’s revisit Agent Plan’s positioning. When it was first released on May 11, Volcano Engine billed it as “the industry’s first Agent package.” This might sound like marketing jargon, but on closer inspection there’s substance.

Traditional model subscriptions sell token quotas or call counts—Cursor and Claude Max follow this path. Agent Plan differs by packaging both Model and Harness (scaffolding/tool chain) together.

Model layer includes an impressive lineup:

• ByteDance in-house: Doubao-Seed series (text/code primary), Doubao-Seedance (video generation), Doubao-Seedream (image generation)
• Third-party leaders: DeepSeek V4 series, GLM-5.1, MiniMax M3, Kimi-K2.6
• Scheduling: manual switching or Auto mode for automatic selection

Harness layer provides essential tools for Agent scenarios:

• Web search (Beta)
• Embedding API
• Vectorization and vision model integration
• ArkClaw intelligent companion

This means that once you connect to Agent Plan, you could write a requirement in Claude Code and have the tools automatically call Seedance 2.0 to generate a video or trigger a web search Skill to fetch the latest data—without having to configure API Keys or write glue code. This is something Cursor hasn’t achieved—Cursor’s multimodal capabilities essentially stop at image input.

AFP Fuel Points: A Transparent Billing Experiment

Agent Plan also introduces AFP (Agent Fuel Points), a unified metric for resources across models and tools.

This design addresses one of developers’ biggest complaints over the past year: how is subscription “usage” calculated exactly? What does Claude Max’s “5x usage” really mean? What’s the difference between Cursor’s “fast request” and “slow request”? Most subscription services are black boxes here.

Volcano’s approach is to convert all model and tool calls into AFP, showing users their exact consumption on the dashboard. This is more developer-friendly, but the real experience depends on whether AFP conversion rules are reasonable—e.g., how many AFP does Seedance 2.0 video generation consume compared to a DeepSeek V4 conversation?

Tool Compatibility: Not Locking Ecosystem Is Smart

Another differentiating point of Agent Plan and Coding Plan is that they don’t bind to a specific IDE or Agent platform. The official list of compatible tools includes:

• Claude Code
• OpenCode
• TRAE
• Hermes Agent
• Cursor, Cline, Kilo Code, Roo Code
• OpenViking

Setup is straightforward, either through the Ark Helper automation tool (MacOS, Linux) or manual Base URL and API Key configuration. Tools compatible with Anthropic and OpenAI protocols can be connected.

This differs entirely from Cursor’s deep binding to its own IDE. Volcano’s logic is: if you’re already comfortable using Claude Code or OpenCode, I’ll fit seamlessly into your workflow—switching the backend model rather than the frontend tool. For developers, migration cost is minimized.

Horizontal Comparison: Key Players in Domestic Coding Subscription Market

After this price cut, the domestic AI Coding subscription landscape is worth reviewing:

1. Volcano Engine Ark Coding/Agent Plan: most diverse models, most open tools, now cheapest
2. Alibaba Tongyi Lingma: deeply bound IDE plugin, Qwen3-Coder led
3. Tencent CodeBuddy: primarily Hunyuan model, bound within Tencent Cloud ecosystem
4. Zhihui, MiniMax and other model vendors’ own subscriptions: mainly single-model

Volcano’s strategy is clear—“I’m not betting on a single model; I’m building the aggregation layer.” This approach is akin to overseas OpenRouter, and some domestic API aggregation platforms: in an era of diverse models, the value of a unified entry point and integrated tool chain grows.

Incidentally, if you need to call GPT, Claude, Gemini, DeepSeek, and other models in your own application, API aggregation solutions like OpenAI Hub also solve the “single Key for all models” problem, with domestic connections and OpenAI format compatibility. They differ from Ark subscriptions—one targets “tool users,” the other “application developers.”

Who Should Get On Board?

Practically speaking, this discount suits three types of users:

Type One: Domestic developers wavering between Claude Pro/Max and Cursor Pro. The ¥9.9/month Lite tier is perfect for a two-month trial. The DeepSeek V4 + GLM-5.1 + Claude Code combo can cover most daily coding scenarios without foreign payment or network issues.

Type Two: Developers already building Agent applications. Agent Plan bundles video generation, image generation, web search, and Embedding—capabilities that otherwise would require separate integrations—making the ¥49.9 Medium tier much cheaper than calling each API individually.

Type Three: Team leaders making technical choices. The ¥200/month Pro tier (¥49.9 for first two months) has 5× the capacity of Lite, enough for a small team’s POC.

Less suitable: heavy commercial production environments. In such cases, it’s better to use Ark model APIs with pay-as-you-go pricing. Subscription capacity, however large, has limits; exceeding them and hitting throttling could affect delivery.

A Quick Assessment

ByteDance chose an interesting timing for this price cut. Early June coincides with the launch of DeepSeek V4, GLM-5.1, MiniMax M3 and other new models, when market sentiment is most receptive—developers are willing to try new things. The ¥9.9 pricing is essentially “customer acquisition cost.” Volcano isn’t looking to recover profit in two months; it’s betting that users will get accustomed to Agent Plan’s workflow and renew at full price from month three.

This approach isn’t new in domestic SaaS history—but in the emerging AI subscription space, Volcano is the first major player to drop the price barrier to this level. Whether Cursor and others will follow suit will be clear in the next two months.

As for Agent Plan’s “Agent Fuel Points” and “Harness tool auto-scheduling” design details, whether they can run smoothly in real-world development will take more developer testing. But for now, the ¥9.9 trial cost is low enough that there’s no reason to hesitate.

References

• From ¥9.9/month, ByteDance Volcano Engine Agent Plan and Coding Plan Limited-Time Discount Announced - IT Home: Official report on the promotion and tier pricing