🎧 Prefer to listen?

$200 a month for Claude Code’s Max plan buys you a phenomenal coding agent. It also buys you vendor lock-in, Anthropic-only models, and code that lives on Anthropic’s servers. I’ve been testing Goose — a free, open-source alternative built by Block and now governed by the Linux Foundation — and the gap between “free” and “$200/month” is smaller than you’d think.

Goose is a command-line AI coding agent. You run it from your terminal, give it tasks in plain English, and it reads your code, edits files, runs commands, and solves multi-step problems. Built in Rust, it connects to over 70 MCP extensions — databases, GitHub, Slack, cloud APIs, browser automation — and supports 15+ LLM providers.

The core difference from Claude Code: Goose is model-agnostic. Point it at Claude, GPT, Gemini, DeepSeek, or a local model running through Ollama. Claude Code only works with Anthropic’s models. This means you can use a cheaper model for simple refactors and a stronger one for architectural work, optimizing cost per task.

Block donated Goose to the Agentic AI Foundation under the Linux Foundation in December 2025, alongside Anthropic’s MCP spec and OpenAI’s AGENTS.md. Vendor-neutral governance — no single company can change the licensing terms or restrict access.

Goose vs Claude Code: Where each one wins

Claude Code handles context windows and multi-file reasoning better than anything else I’ve tested. Working across a large codebase, its ability to map relationships between files is genuinely impressive. If your day is pure code — reading, writing, refactoring across dozens of files — Claude Code still has an edge.

Goose competes on most coding tasks but has a different strength profile. Where Claude Code goes deep on code reasoning, Goose goes wide. Through its MCP extension ecosystem, it reaches into databases, APIs, Slack, cloud services, and browser automation. If your work involves deploying, managing infrastructure, or automating workflows beyond editing code, Goose covers more ground.

On raw coding benchmarks, the gap has narrowed. We covered [NousCoder-14B](/posts/nouscoder-14b-op