Baidu Qianfan Coding Plan Stops Renewal: A Short-Lived 4-Month Experiment

Today (June 25), the Baidu Qianfan team released an announcement: All Coding Plan packages will stop renewal starting immediately.

This AI coding subscription service, launched only this February, lasted just 4 months. Users who have purchased it can continue using it until their current service period ends; the auto-renewal feature will be disabled in 30 days. After that, the product officially becomes history.

Review: What Coding Plan Did

When released in February, Baidu Qianfan positioned Coding Plan with three selling points:

• Multi-model integration: Mainstream coding models like GLM-4.7, DeepSeek-V3.2, switchable with one click on the console
• High compatibility: Standardized interfaces, directly compatible with tools like Claude Code and Cursor
• Low-cost subscription: Lite version: first month 9.9 RMB (then 40 RMB/month), Pro version: 49.9 RMB/month

The positioning was clear—an “AI coding intermediate layer” that allows developers to use multiple models with a single subscription, without needing to integrate each API themselves or manage multiple keys.

The idea itself was sound. At the time, AI coding solutions on the market either bound you to a single model (e.g. Cursor defaults to Claude) or charged on usage (pay as you go). Coding Plan tried to lower developers’ decision-making costs with a fixed monthly fee + multiple models.

Problems Started as Early as March

But the good times didn’t last. Only a month after launch, Baidu Qianfan issued a “Stagewise Resource Adjustment Announcement”:

Due to continuing popularity of the Baidu Qianfan Coding Plan subscription, model API call volume has sharply increased.

The announcement stated two changes:

1. First-purchase discount stopped: The original 9.9 RMB first-month discount stock ran out and would no longer be offered
2. Peak-period throttling: Some models would have their calls restricted during peak hours

The official explanation was “resource integration and expansion optimization,” suggesting users switch to other models during throttling.

The wording here is subtle. “Subscription popularity” sounds like good news, but for cloud services, too many users with scarce resources often means structural cost issues.

The risk of a fixed monthly fee is precisely this: If users’ actual call volume far exceeds expectations, the provider loses money. And inference costs for AI models aren’t cheap—especially large-parameter models like GLM-4.7.

From Throttling to Shutdown: What Happened in 3 Months

March throttling, June shutdown. In between, the official team released no product updates or feature iterations.

We can make some reasonable guesses:

Cost Pressure

The 49.9 RMB/month Pro version offered 90,000 request credits. Based on current mainstream model API pricing, this rate was practically “subsidizing users to acquire them.” If users really maxed out the credits, Baidu would likely be losing money.

AI coding scenarios by nature require high call frequency—writing, debugging, optimizing code can mean dozens or hundreds of calls a day per developer, unlike casual chat apps where users converse briefly and leave.

Imbalance Between User Growth and Resource Expansion

Based on the March announcement, user growth exceeded expectations. But cloud resource expansion, especially GPU compute power, is not something that happens overnight.

Baidu Intelligent Cloud certainly has substantial compute reserves, but for a low-priced subscription product like Coding Plan, how much resource allocation it could secure is another matter.

Intensifying Market Competition

Over these four months, the AI coding sector saw many changes:

• Domestic major models all boosting coding capabilities
• Cursor, Windsurf, and other tools steadily growing user bases
• Pay-as-you-go API aggregation services maturing quickly

Coding Plan’s “multi-model + low-price subscription” approach failed to build sufficient competitive barriers.

Three User Groups Affected

According to the announcement, this adjustment affects three types of users:

| User Type | Impact | Handling | |-----------|--------|----------| | Existing package users | Can continue using until service period ends | No action needed | | Auto-renewal enabled users | Auto-renewal disabled after 30 days | Recommended to monitor expiry date | | Expired package users | Cannot renew | Need to migrate to another plan |

If you are a current Coding Plan user, you now need to consider migration options.

Migration Option 1: Use Model’s Official API Directly

If you mainly use one model (e.g. DeepSeek-V3.2), you can apply directly for an official API Key. Pay-as-you-go ensures costs are controllable.

Migration Option 2: Use an API Aggregation Platform

If you need flexibility in switching models, consider API aggregation services. These platforms allow calling multiple models with a single key, support OpenAI-compatible interfaces, and have low migration costs.

For example, OpenAI Hub supports mainstream models like GPT, Claude, Gemini, DeepSeek, offers direct access in China, and charges per use. If your Cursor or Claude Code configuration was OpenAI-compatible, switching is basically just updating the Base URL and API Key.

Migration Option 3: Use IDE Built-In AI Features

If your needs are simple, many IDEs now integrate AI coding. For example, VS Code’s Copilot plugin, JetBrains’ AI Assistant. While model selection is less flexible, they work out-of-the-box and are convenient.

Some Observations

1. Fixed Monthly Subscriptions Are Hard to Sustain in AI

AI model inference costs are tangible—unlike traditional SaaS, where marginal costs are near zero. The more users use, the more the provider spends.

This creates a dilemma for fixed-monthly pricing:

• Price too low—users max out quotas and provider loses money
• Price too high—users won’t pay

Coding Plan chose low pricing to attract users but failed to balance user growth with resource supply.

2. “Multi-model integration” Is Not a Moat

Multi-model switching is a good feature but has low technical barriers. Any API aggregation platform can do something similar.

Coding Plan’s real value should have been in the “integrated experience”—letting developers ignore which underlying model is running and just code. But user feedback indicated the experience wasn’t good enough. Throttling and switch delays increased usage costs instead.

3. The Dilemma of Cloud Vendors Making Developer-focused Products

Baidu Intelligent Cloud’s main business is enterprise services. Low-priced developer-targeted subscriptions may have low internal priority.

When growth and cost conflict, enterprises often choose to cut the product and focus resources on more commercially valuable directions.

Advice for Developers

If you are using or considering AI coding tools, here are some tips:

Don’t Over-rely on a Single Service

Today it’s Coding Plan shutting down, tomorrow it could be another product. The AI field changes too fast; any service could shift strategies.

Build core workflows on standardized interfaces—like OpenAI-compatible formats—so switching providers is just a configuration change.

Pay-as-you-go May Be Cheaper than Subscription

Fixed monthly prices seem simpler, but if your usage is volatile, pay-as-you-go may be more flexible. Low-use months cost nothing, high-use months cost more, but overall may be lower.

Focus on Model Capabilities, Not Packaging

The core of an AI coding tool is the underlying model’s capability. Rather than focusing on how the product is packaged, directly evaluate the model—code generation quality, contextual understanding, response speed.

Today’s mainstream coding models (Claude, GPT-4o, DeepSeek-V3.2, GLM-4.7, etc.) can all be called via API. The tool is just a shell.

In Closing

Baidu Qianfan Coding Plan’s 4-month lifespan is a microcosm of China’s AI coding sector.

Low-price strategies can quickly acquire users but can’t solve fundamental business model issues. When growth conflicts with costs, a product’s fate is already determined.

For developers, this is a reminder: while enjoying the benefits of AI tools, be prepared for possible changes. Workflows built on standard interfaces are more reliable than relying on any specific product.

---

References

• Baidu Qianfan Coding Plan Stops Renewal: Purchased users can continue until service expires – IT Home: Detailed coverage of the official announcement, including handling methods for three types of users