Claude Code, Anthropic's AI coding agent, costs up to $200 per month, sparking developer backlash. Goose, an open-source alternative from Block, offers similar functionality for free with local execution and no rate limits, gaining over 26,100 GitHub stars.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Anthropic's Claude Code pricing sparks developer backlash, while Block's open-source Goose offers a free, local, no-limit alternative that quickly gains community traction.
Anthropic's Claude Code pricing has sparked strong developer backlash. The free plan offers no access; the Pro plan (annual $17/month or monthly $20) provides only 10-40 prompts per 5 hours; the Max plan at $100-200/month offers 50-200 and 200-800 prompts respectively, and includes the most powerful model Claude 4.5 Opus.
In late July 2025, Anthropic introduced new weekly rate limits: Pro users get 40-80 hours of Sonnet 4 per week, while $200 Max users get 240-480 hours of Sonnet 4 plus 24-40 hours of Opus 4. However, these 'hours' are actually token limits; independent analysis shows Pro users effectively get about 44,000 tokens, and $200 Max users about 220,000 tokens. Developers criticize the limits as 'vague and confusing,' with some users reporting exhausting their daily quota within 30 minutes.
Anthropic argues that the limits affect fewer than 5% of users and target those running '24/7,' but it hasn't clarified whether that percentage refers to Max users or all users. This ambiguity has intensified community dissatisfaction, with some users canceling subscriptions.
Goose, developed by Block (formerly Square, led by Jack Dorsey), is a 'local AI agent' that runs entirely on the user's machine using open-source models. Its core selling points: no subscription fees, no cloud dependency, no rate limits, data stays local, and it can even be used offline on a plane.
Goose has rapidly gained popularity on GitHub, amassing over 26,100 stars, 362 contributors, and 102 releases, with the latest version 1.20.1 released on January 19, 2026. Its documentation describes it as 'beyond code suggestions,' capable of 'installing, executing, editing, and testing any LLM.'
Goose is model-agnostic: it can connect to Claude (via API), OpenAI GPT-5, Google Gemini, or run open-source models like Llama, Qwen, DeepSeek locally via Ollama. In local setups, there are no usage limits whatsoever.
Goose, available as a command-line tool or desktop app, can autonomously perform complex development tasks: building projects from scratch, writing and executing code, debugging, orchestrating multi-file workflows, and calling external APIs. Its architecture relies on 'tool calling,' where the language model requests external systems to perform specific actions.
Tool-calling performance depends on the underlying model. According to the Berkeley Function Calling Leaderboard, Anthropic's Claude 4 model currently performs best, but open-source models (such as Llama, Qwen, Gemma, DeepSeek) are rapidly catching up. Goose also integrates the Model Context Protocol (MCP), enabling connections to databases, search engines, file systems, and third-party APIs.
For a completely free and privacy-preserving setup, developers need to install Ollama (for running models locally) and Goose, then pull a model like Qwen 2.5. The entire process requires no internet connection, and data never leaves the local machine.
This article is based on a VentureBeat report, which includes references to developer forum feedback and independent analysis, but some data (e.g., specific token limit values) comes from independent analysis rather than official Anthropic confirmation. Goose's GitHub data (stars, contributors, releases) comes from public repositories and is verifiable. Anthropic's statement (affecting fewer than 5% of users) is a source claim, with the user scope unspecified.
Goose directly challenges Claude Code's high pricing and restrictive model with its free, open-source, local, and model-agnostic approach, but tool-calling performance still depends on model choice, and open-source models may fall short of Claude 4 on complex tasks.
Primary report
Primary source