Core Update Summary

FocusAny, an open-source large language model platform, has released version 2.2.0, focusing on developer experience and platform efficiency. This update involves no pricing changes or weight openness adjustments and is free for existing users. Key changes include:

• Model capability labels launched: Intuitive capability tags added next to each model in the model settings page
• SDK CLI enhanced: New diagnose and forward commands added
• Plugin resource optimization: Plugin developers no longer need to bundle large files; resources are downloaded on-demand

Model Selection Experience Upgrade: Capabilities Displayed Clearly

Previously, users had to manually discern available functions among dozens of model names in the model settings, resulting in high decision complexity. Version 2.2.0 introduces capability tags, clearly indicating each model’s supported core features:

• Image input support (“Visual”)
• Tool calling capability (“Tool use”)

This design directly addresses developer feedback—information transitions from “hidden configuration” to “immediately visible”, significantly improving model comparison efficiency. In previous versions, users needed to enter model detail pages or consult documentation to confirm basic capabilities; now, initial screening can be completed with a quick visual scan.

This enhancement is a “zero-intrusion optimization”: it does not alter underlying model execution logic, only front-end information presentation. It benefits scenarios involving mixed-model usage, such as task pipelines combining visual understanding and logical reasoning models.

SDK CLI Enhancement: Diagnostics and Forwarding Implementation

As a common entry point for developers integrating with FocusAny, the SDK CLI now includes two critical features:

1. diagnose command: Quickly detects current development environment configuration issues, including network connectivity, authentication status, and SDK version compatibility
2. forward command: Supports forwarding requests to specified models or proxy addresses, facilitating debugging and gray-scale testing

These additions reduce local debugging barriers. For instance, developers can use forward to point to pre-release models for logic validation while simulating production environments locally, then use diagnose to ensure environmental correctness, avoiding repeated submissions to production for troubleshooting.

Plugin Development Workflow Reengineering: On-Demand Resource Download

Plugin developers告别 the old “bundle large files” model. Previously, to ensure plugin package independence, developers had to bundle resource files (such as model weights, corpora) together, causing plug-in installation packages to often exceed hundreds of MB. Version 2.2.0 allows plugins to declare required resources, with the platform dynamically downloading based on actual usage scenarios:

• Significantly reduced plugin package size, improving distribution efficiency
• On-demand download during first use of specific resources, avoiding unnecessary storage occupation
• Shared resources among multiple plugins, saving bandwidth and disk space

Key comparison: One plugin previously bundled 280MB audio processing resources; post-update, its installation package is reduced to 12MB, dynamically downloading only required modules on first call.

Implementation Recommendations

• Upgrade immediately if: Your project integrates multiple models, frequently debugs via SDK CLI, or develops plugins; capability labels provide the most significant efficiency gains for mixed-model applications
• Wait for now if: Your application uses static models without CLI tool usage or plugin resource dependencies; this update offers limited benefits and can be deferred until subsequent feature requirements align