What each one says about the other
These sentences were written about each tool on its own, before this page existed, and each names the other tool by name. They are quoted, not composed.
Claude Code on Crush · not for
A developer who must run a local or self-hosted model should choose goose or Crush; both accept an Ollama endpoint.
“Local model runner supporting Qwen, Llama, DeepSeek, and other open-source models.”
Claude Code and Crush, fact by fact
| Fact | Claude Code | Crush |
|---|---|---|
| StarsClaude Code has 121,307 more stars (5.3×). | 149,849Verified 8 Oct 2026 | 28,542Verified 8 Oct 2026 |
| Newest commitBoth committed on the same day. | 8 Oct 20260 days old when we read itVerified 8 Oct 2026 | 6 Oct 20261 days old when we read itVerified 8 Oct 2026 |
| ContributorsCrush has 97 more contributors (2.7×). | 57Verified 8 Oct 2026 | 154Verified 8 Oct 2026 |
| License | not tracked for Claude Code | OtherVerified 8 Oct 2026 |
| Latest stable releaseBoth released on the same day. | v2.1.294published 8 Oct 2026Verified 8 Oct 2026 | v0.97.1published 29 Sep 2026Verified 8 Oct 2026 |
| npm downloadsClaude Code has 14,778,510 more npm downloads a week (1902×). | 14,786,283 a week@anthropic-ai/claude-codeVerified 8 Oct 2026 | 7,773 a week@charmland/crushVerified 8 Oct 2026 |
| Repository | anthropics/claude-codeTypeScriptVerified 8 Oct 2026 | charmbracelet/crushGoVerified 8 Oct 2026 |
What we know about Claude Code
Claude Code runs only Anthropic models: every alias — default, best, fable, sonnet, opus, haiku and opusplan — resolves to a Claude model, and changing the base URL moves where the request goes without changing which model answers it.
“ANTHROPIC_BASE_URL changes where requests are sent, not which model answers them.”
Plan a change with Opus and execute it with Sonnet inside one session, by setting the model to `opusplan`.
“Special mode that uses opus during plan mode, then switches to sonnet for execution”
Run the agent on infrastructure your employer already approved: the model setting accepts an Amazon Bedrock inference profile ARN, a Microsoft Foundry deployment name, or a Google Cloud Agent Platform version name.
“Amazon Bedrock: an inference profile ARN”
Work one long task without losing history: `sonnet[1m]` gives a 1 million token context window, and auto-compaction runs below it.
“Uses Sonnet with a 1 million token context window for long sessions”
You cannot audit, fork or patch the agent: the repository publishes the changelog, plugins and examples, not the CLI source, under an all-rights-reserved licence. GitHub Copilot CLI is the only other tool in this set that does the same.
“© Anthropic PBC. All rights reserved. Use is subject to Anthropic's [Commercial Terms of Service](https://www.anthropic.com/legal/commercial-terms).”
Subscribers report the plan limit arriving far earlier than the plan implies: the repository's most-commented open issue carries 1,487 comments and 724 reactions and has stayed open since 2026-01-03.
“[BUG] Instantly hitting usage limits with Max subscription”
On API-key billing the cost is per token, and Anthropic's own enterprise figure is $150-250 per developer per month.
“Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users.”
A developer who must run a local or self-hosted model should choose goose or Crush; both accept an Ollama endpoint.
“Local model runner supporting Qwen, Llama, DeepSeek, and other open-source models.”
What Claude Code costs
Claude Pro costs $17 per month on the annual plan, $200 billed up front, or $20 per month billed monthly; Max starts at $100 per month.
“Per month with annual subscription discount ($200 billed up front). $20 if billed monthly.”
Pro and Max subscribers do not pay per token for Claude Code; the usage sits inside the subscription.
“Claude Max and Pro subscribers have usage included in their subscription, so the session cost figure isn’t relevant for billing purposes.”
What we know about Crush
Crush is configured by a `crushrc` that is Bash, and it runs that Bash identically on Windows because Crush embeds its own Bash interpreter rather than calling the host shell.
“Because Crush has a native, built-in Bash interpreter, Bash-based config works identically across all platforms, including Windows.”
Code in a terminal on a platform the other agents skip: Crush ships first-class builds for Android, FreeBSD, OpenBSD and NetBSD as well as macOS, Linux and Windows.
“- **Works Everywhere:** first-class support in every terminal on macOS, Linux, Windows (PowerShell and WSL), Android, FreeBSD, OpenBSD, and NetBSD”
Work in a typed language where the compiler knows more than the file: Crush attaches language servers, so `lsp add go --command "gopls"` feeds gopls output into the agent's context.
“- **LSP-Enhanced:** Crush uses LSPs for additional context, just like you do”
Start a task on an expensive model and finish it on a cheap one: Crush swaps the model mid-session and keeps the context.
“- **Flexible:** switch LLMs mid-session while preserving context”
Crush is not open source today: it ships under the Functional Source License 1.1, which bars any competing commercial use until each version turns MIT two years after release.
“A Permitted Purpose is any purpose other than a Competing Use. A Competing Use · means making the Software available to others in a commercial product or · service that: · 1. substitutes for the Software;”
A new model reaches Crush only when a separate repository merges it, because the default model list lives in Catwalk, not in Crush.
“Crush’s default model listing is managed in [Catwalk](https://github.com/charmbracelet/catwalk), a community-supported, open source repository of Crush-compatible models, and you’re welcome to contribute.”
Crush has no read-only planning mode: the request for Build and Plan modes has been open since 2025-12-28 with 61 reactions.
“Build / Plan Modes”
A team that needs an approval prompt on every shell command out of the box should pick Codex CLI or GitHub Copilot CLI.
“- **Full control:** Preview every action before execution — nothing happens without your explicit approval.”
What Crush costs
Charm Hyper, the provider built for Crush, gives 100 Hypercredits a month at $0 and 250 a day on a $20 monthly subscription; 1 Hypercredit is 5 cents.
“Hypercredits are the core of Hyper's usage-based pricing model, where credits map directly to token costs. 1 Hypercredit is currently 5¢, and they're distributed based on your subscription tier.”
Questions people ask
- Are Claude Code and Crush still maintained?
- Claude Code committed to anthropics/claude-code (main) on 8 Oct 2026. Crush committed to charmbracelet/crush (main) on 6 Oct 2026. Both readings come from the default branch of each repository, on the date shown beside them.
- What do Claude Code and Crush cost?
- Claude Code: Claude Pro costs $17 per month on the annual plan, $200 billed up front, or $20 per month billed monthly; Max starts at $100 per month. Pro and Max subscribers do not pay per token for Claude Code; the usage sits inside the subscription. Crush: Charm Hyper, the provider built for Crush, gives 100 Hypercredits a month at $0 and 250 a day on a $20 monthly subscription; 1 Hypercredit is 5 cents. Every figure here is quoted from the vendor's own page and linked beside the claim.
- Is Claude Code or Crush open source?
- We have not read a licence for Claude Code. Crush's repository charmbracelet/crush reports the Other licence. We read the licence GitHub reports for the repository, which is not the same question as whether the hosted product is free.
How we check this page
- Machine facts. Stars, contributors, the newest commit on the default branch, the licence GitHub reports and the latest stable release come from the GitHub API, and download counts from npm, PyPI, Homebrew and the VS Code Marketplace. Each value carries the day it was last confirmed, never the day this page was built. The line under each label is arithmetic on those two values.
- Judgments. Every claim about what a tool suits, what it is bad at and what it costs is quoted from a page we read, and that page is linked beside the claim. Where there is no link, we wrote no claim.
- Index status. Both Claude Code and Crush carry claims we read, quoted and linked. So this page is offered to search engines.
Related comparisons
Pairs we can answer for come first.
