Droid CLI + GLM-4.6 + FREE 4.5 Sonnet,GPT-5 : Is it really the NEXT CLAUDE CODE?

Key Concepts

• Droid: A CLI tool for agentic coding assistance.
• Agentic Performance: The ability of an AI model to autonomously perform complex tasks.
• Sub-agents: Smaller, specialized agents within a larger agent system.
• Ninja Chat: An all-in-one AI platform offering access to various AI models.
• GLM 4.6: A specific AI model used for coding.
• Radix UI: A React UI component library.
• Expo: A framework for building React Native applications.
• Godo: A free and open-source game engine.
• Open Code Repo: A large, existing code repository used for testing.
• Tool Calls: The ability of an AI model to use external tools or APIs.

Droid Overview and Setup

• Droid is a CLI tool that allows users to leverage agentic coding capabilities with various AI models.
• While not open source, the CLI tool is free, requiring users to set up their own API key and base URL.
• Droid offers a free pro trial without requiring a credit card, similar to Cursor or Windsurf, providing access to models like Sonnet and GPT-5 with a token allowance.
• The tool boasts agentic performance that can elevate Sonnet's capabilities to match Opus.
• Installation is done via a command-line command, followed by account setup.
• Droid supports sub-agents, similar to Claude Code, with a comparable defining process.
• IDE integration is available, reminiscent of the earlier Claude Code extension, allowing code chunk referencing and workspace awareness.
• Custom droids (sub-agents) are supported, potentially allowing for easy porting of Claude Code sub-agents.
• The tool supports the agents markdown specification and allows users to bring their own provider and key.
• MCP (Multi-Chain Processing) is also supported.

Ninja Chat Advertisement

• Ninja Chat is presented as an all-in-one AI platform with access to models like GPT-4o, Claude for Sonnet, and Gemini 2.5 Pro for $11 per month.
• Features include an AI playground for comparing model responses and a mind map generator.
• The basic plan includes 1,000 messages, 30 images, and 5 videos monthly, with higher tiers available.
• Discount codes "king25" (25% off any plan) and "king40yearly" (40% off annual subscriptions) are provided.

Droid Interface and Features

• The Droid interface features a large branding at the top and a prompt box at the bottom.
• The default model is GPT-4.5.
• Slash commands are available, including:
  • /model: To switch between models like Sonnet High, GPT-5, and Codex.
  • /cost: To track token usage.
  • /settings: To adjust model, reasoning effort, diff display mode, completion bell, cloud sync, and custom droids.
• Most features found in Claude Code are also available in Droid.

Performance Evaluation with Sonnet 4.5

• A benchmark question was used: creating a movie tracker app using Expo.
• The process is similar to Claude Code, but Droid doesn't ask for approval for edits.
• The generated app had issues, including using Radix UI without installing the package.
• The generated project was considered ordinary compared to Baseclaw code or Open Code.
• The calendar UI was good, but there were issues with font colors.
• A Go TUI calculator task failed.
• Editing a Godo FPS game to add a step counter and life bar also failed. The life bar was implemented, but the step counter didn't work.
• The tool struggled to "one-shot" tasks.
• Testing with a large open code repo also failed.
• The performance was deemed below GPT-5 Codex.
• The UI was liked for its snappiness and similarity to Claude Code.
• The tool seemed to perform multiple tool calls at once.

Performance Evaluation with GLM 4.6

• GLM 4.6 was tested to see if Sonnet was the bottleneck.
• The API was configured in the CLI settings file located at a specific path.
• GLM 4.6 worked well without issues and was fast.
• The movie tracker app generated with GLM 4.6 was light-themed and included a calendar.
• The calculator task was good, comparable to Claude Code.
• The Godo example worked fine, with the life bar and step counter functioning correctly.
• The open code repo question still failed.
• While promising, GLM 4.6 didn't significantly increase the model's raw capabilities.

Conclusion

• Droid doesn't bring anything new to the table and is not recommended over Claude Code, Kilo Code, or similar tools for general coding.
• The presenter expressed disappointment with the tool's performance, especially given the hype and claims of top-scoring agentic performance.
• The UI is considered cool.