Claude Pro, Kagi Assistant and Charm Hyper

I'll pull the source article so the paragraphs stick to real details rather than inventing numbers or product claims.A subscriber has cancelled **Claude Pro** while still treating Claude as the leading general-purpose and coding model. The decision was about redundant spend: the writer already pays for **Kagi** (upgraded to the **Duo** plan) and gets **Kagi Assistant** with that subscription. On general consolidated-search questions, side-by-side checks found Claude and Kagi Assistant answers often similar or the same.

Coding use of **Claude Code** is light—about once a week for tasks faster than rusty manual chops. All of the writer’s servers run **FreeBSD**. Claude Code dropped FreeBSD support after Anthropic moved distribution from **npm** to shipping its own package (tracked in GitHub issue **#30640**). That platform cut makes a paid Claude coding seat harder to justify for a FreeBSD-heavy setup.

The replacement coding path is Charm’s stack. Longtime use of Charm’s **mods** led to **Crush**, a Go CLI coding agent. Charm has extended Crush with **Hyper** (hyper.charm.land), a subscription inference service. Crush is a Go binary rather than a heavier client stack: it feels snappier than Claude Code and is LLM-agnostic—you supply API tokens. Hyper adds access to lower-priced models the writer finds often close enough to Anthropic’s for day-to-day work.

For engineers and builders, this is a stack re-split by job, not a model brand swap. Search-style answers ride the search subscription; coding rides an agnostic CLI plus optional cheap inference; frontier Claude is kept only when quality justifies burning tokens. Platform support (FreeBSD vs package strategy) and subscription overlap matter as much as raw model quality when weekly coding volume is low.

Market-wise, the post sits in the same HN Claude/Codex tooling conversation but is a personal cost-and-workflow note, not a benchmark suite. **Hyper** does not yet expose **Kimi K3**, which the piece flags as a serious frontier contender expected to land on Hyper soon given Charm’s pace on Crush and Hyper. Claude stays in the mix via buffered Anthropic tokens in Crush or chat in **TypingMind**, with Kagi Assistant covering the rest.

Practical takeaway: audit whether you are paying twice for answer-style LLM use if Kagi (or similar) already includes an assistant; check whether your agent’s install path still covers your OS; and prefer an LLM-agnostic CLI when you want to mix cheap inference with occasional Anthropic tokens. Watch Hyper’s model catalog (especially Kimi K3) and Claude Code’s packaging/platform support before renewing a Pro seat you barely use for coding.