Anthropic just blocked OpenClaw. Here’s what you need to do immediately

Summary

In this video, the creator explains how to continue using OpenClaw effectively despite Anthropic banning its subscriptions. They introduce a cost-efficient system that leverages Opus 46 as the orchestration 'brain' while using cheaper or alternative models as task-executing 'muscles.' The setup covers coding, research, writing, and web tasks, recommending specific models like ChatGPT 5.4, GLM 5.1, and Quen 35 depending on hardware and subscription tiers. Step-by-step instructions for integrating these models into OpenClaw are provided, emphasizing reliability, performance, and ROI. The video highlights strategies for maximizing AI efficiency and savings while maintaining top-tier task completion.

Takeaways

• 😀 Anthropic has banned OpenClaw from all its subscriptions, making it impossible to use the $200 or $20 subscriptions for OpenClaw integration.
• 😀 OpenClaw now requires users to pay via the API, which charges per usage. This can become very expensive, especially with heavy use.
• 😀 A workaround system has been developed to continue using OpenClaw and Opus while saving money, utilizing Opus for orchestration and cheaper models for task execution.
• 😀 The key to saving money with OpenClaw is to use a powerful orchestrator like Opus for decision-making, and delegate tasks to less expensive models.
• 😀 Opus 46 is recommended as the top orchestrator for OpenClaw, providing unparalleled task completion reliability compared to other models like ChatGPT.
• 😀 Using ChatGPT for coding tasks can save money while ensuring good quality execution, especially if using the $20/month or Pro-tier subscription for access to more coding capabilities.
• 😀 For tasks beyond orchestration and coding (like research and writing), cheaper models such as GLM 5.1 or even Google’s AI services can be used to reduce costs.
• 😀 Even with basic hardware (e.g., a Mac Mini), users can set up OpenClaw with cloud-based models and still achieve powerful performance at an affordable cost.
• 😀 Those with good hardware (e.g., Mac Studios or DJX Sparks) can run local models like Quen 35 and Gemma 4 for coding and other tasks, offering unlimited usage without the high API costs.
• 😀 The final setup includes configuring OpenClaw to integrate multiple models for different tasks, such as using Opus for orchestration, ChatGPT for coding, and other models for specific tasks like writing or web scraping.
• 😀 Despite the price increase with API-based payments, using Opus and this new system is still considered a great deal compared to hiring a full-time employee, offering strong ROI for digital AI assistance.

Q & A

Why did Anthropic ban OpenClaw from all subscriptions?

Anthropic has banned OpenClaw from all subscriptions, meaning users can no longer connect it to their $200 or $20 subscriptions. The only option left is to pay for the API, which is more expensive since you pay for every single use.

What is the major issue with using OpenClaw's API?

The issue with OpenClaw’s API is that it charges for each use, meaning if you use OpenClaw frequently, it can result in hundreds or even thousands of dollars in costs per day, making it unsustainable for heavy users.

How does the orchestrator model help save costs?

The orchestrator model, specifically using Opus 46, allows you to handle tasks efficiently by using cheaper models for specific tasks like coding and web search. This reduces overall costs while still maintaining high performance and quality.

What makes Opus 46 the best model for orchestration in OpenClaw?

Opus 46 is considered the best for orchestration due to its ability to reliably complete tasks. It can effectively manage task execution and validate their completion, making it the most reliable and powerful option for running OpenClaw.

Why is Opus considered better than other models like ChatGPT for OpenClaw?

Opus is specifically optimized for task completion and orchestration in OpenClaw, whereas models like ChatGPT are not as reliable in finishing tasks. Many other models, including ChatGPT, often fail to complete tasks consistently.

What is the recommended setup for users with average hardware?

For users with average hardware, the recommended setup is to use Opus 46 as the orchestrator and rely on cheaper models like ChatGPT 54 (for coding) and cheaper options like GLM 5.1 or Google models for tasks like writing, web searching, or research.

What should users with high-end hardware consider for their OpenClaw setup?

Users with high-end hardware, like a 512GB Mac Studio, should consider running local models such as Quen 35 for coding and Gemma 4 for tasks like web scraping. This provides unlimited usage and can unlock more use cases, though it requires more advanced hardware.

How can local models like Quen 35 improve the setup for high-end hardware users?

Local models like Quen 35 allow users with high-end hardware to have unlimited usage for tasks like coding. This is particularly beneficial for those who need continuous processing and prefer not to rely on cloud-based models for coding tasks.

What is the advantage of using cheaper models like GLM 5.1 for certain tasks?

Using cheaper models like GLM 5.1 for tasks such as writing, research, and web scraping allows users to save on API costs while still maintaining a reasonable level of performance. GLM 5.1 is a budget-friendly option that performs well for general tasks.

How do discounted API bundles help users save money with OpenClaw?

Purchasing discounted API bundles from Anthropic can save users money by offering lower rates per token. The more credits you buy upfront, the greater the discount, which is ideal for users who plan to use OpenClaw heavily over time.