Cursor's Series of "Mishaps": Model Unavailability and Pricing Issues

Cursor's successive "mishaps" (model regional blockades + pricing halved-style adjustments) are just the tip of the iceberg. Meanwhile, the wildly popular Manus was forced to "withdraw from China and relocate to Singapore" as it could no longer obtain authorization from its US suppliers, marking a further tightening of geopolitical regulation's grip on AI developers. Competitors like Codeium, Claude Code, Augment Code, and Copilot are frantically releasing new features, shattering Cursor's price advantage and functional differentiation.

1. Regional Blockade: The "Supply Cutoff Crisis" from Cursor to Manus

1.1 Cursor: Chinese Users Repeatedly Encounter "Model not available"

Since July of this year, domestic developers frequently receive the following error when calling GPT-4o or Claude Sonnet in Cursor:

Model not available — This model provider doesn’t serve your region

There are two reasons:

Supplier Compliance Tightening: OpenAI, Anthropic, etc., have added "restricted regions" clauses to their API TOS, directly listing China.
HTTP/2 Compatibility: Cursor defaults to HTTP/2 for streaming output. Many corporate proxies/campus networks are not friendly to this protocol; a handshake failure can be misjudged as a "regional blockade."

Solutions:

Set Preferences → Network → HTTP Compatibility Mode = HTTP/1.1 or 1.0, forcing requests to fall back to chunked transfer.
If your VPN supports TUN Mode, route 0.0.0.0/0 through the VPN to bypass application-layer blocking.

1.2 Manus: The "Upgraded Version" of Compliance Blockade

On July 11th, multiple media outlets confirmed:

AI assistant Manus, under Beijing Butterfly Effect Technology, decided to conduct large-scale layoffs of its Chinese team and relocate entirely to Singapore due to US export controls and model licensing issues.
The Chinese version of Manus, developed in cooperation with Alibaba, has stopped iterating, and related social media content has been completely cleared.
Experts speculate that Manus initially called Claude 3.5 Sonnet. According to US export regulations, this constitutes "unauthorized model supply" within mainland China.

Implication: When an "AI grunt work" end-user product cannot secure its upstream model supply chain (or meet its regional compliance requirements), even a massive market can be "zeroed out overnight."

2. Pricing Out of Control: The End of Cursor's 500-Request Myth

Time	Event	Outcome
2025-06-16	$20 Pro = 500 requests changed to token-based billing, introducing $200 Ultra (20× quota)	Many users faced "sudden fee hikes"
2025-07-07	CEO publicly apologized, admitting "insufficient communication"	TechCrunch reported refund announcement
2025-07-14	Official blog: Full refunds for excess usage between June 16 – July 4	Cursor Blog announcement

2.1 Why the Inevitable Switch to "Token-Based" Billing?

Cost per request can vary by 100×: small completions ≈ $0.01; large-context reasoning ≈ $1+.
Within the same "500-request" bucket, light users subsidize heavy users—a "ticking time bomb" for any SaaS.

2.2 Why Are Users Still Angry?

Lack of warning: Bills spiked without a real-time usage dashboard.
Low switching cost: VS Code extensions allow one-click source switching; Codeium/Copilot are immediately available.

2.3 User Coping Strategies

Revert to "Fixed Bucket" Mode:

Turn off usage-based pricing in Cursor settings to continue enjoying the 500-request limit.
Before month-end, use up the quota and register a second Pro account for $20 to get another 500 requests.

3. Competitor Siege: Free Tiers, Hybrid Reasoning, and Deep Integration

Product	Core Advantage	Price/Strategy	Key Developments
Codeium	70+ languages + Chat + Repo Q&A	Free, unlimited	Reddit reputation skyrocketing
Claude Code CLI	Hybrid Reasoning, explicit "chain-of-thought"	$50+/month	Anthropic released 3.7 Sonnet
Augment Code	JetBrains & VS Code dual ecosystem; static analysis integration	Gartner rating 4.8	Official evaluation page
GitHub Copilot	Deep IDE integration + enterprise private-domain knowledge	$10–19/seat	Launched new features like Copilot Chat

Current Market Share Erosion:

Gartner Peer Insights comparison: Augment 4.8 ★ vs Cursor 4.5 ★
In a Medium review of 12 tools, Codeium was rated "best free option"

Cursor's original moat of "$20 for 500 requests, all features included" is being shattered by free tiers and multimodal hybrid reasoning.

4. Checklist for Developers and SaaS Entrepreneurs

4.1 For Developers

Dual-network insurance: Switch to HTTP/1.x within the IDE; externally use a VPN supporting TUN mode.
Cost visualization: Write scripts to pull from the Cursor Usage API and set up Grafana alerts.
Multi-source toolchain: Primary ↔ Codeium/Copilot, backup ↔ Claude Code + KIMI K2, local ↔ Llama 3 + ollama.

4.2 For Entrepreneurs

Risk	Defensive Measures
Regional Compliance	Multi-cloud, multi-model: Connect >2 model providers for the same feature; self-host inference for sensitive regions
Billing Trust	Real-time dashboard + soft-limit reminders: Pop-up at 80% threshold; hard-limit freeze at 90%
Competitor Price War	Differentiate: Enterprise private knowledge bases, private deployment, deep process coupling, not just basic completion

For Cursor's settings, you still need to configure them

5. Conclusion: The Next Stage of AI Tool Moats

The "successive Waterloo moments" of Cursor and Manus serve as a wake-up call for the industry: Upstream model compliance, transparent and predictable billing, and a multi-source strategy that allows for switching at any time will become critical for the survival of AI tools post-2025. Developers must learn to "walk on multiple legs," and startup teams must design business models that can withstand the dual extremes of "the heaviest users + the strictest regulations." Otherwise, a single regional blockade or pricing misstep can instantly wipe out a product's entire moat.

Reference Sources

TechCrunch: Cursor apologizes for unclear pricing changes — TechCrunch's report on the Cursor CEO's public apology for the unclear pricing changes in June 2025, which details the refund controversy timeline referenced in the article's table.