WOZCODE claims 54% lower Claude Code cost on Opus 4.7 benchmark
Users posted WOZCODE, a Claude Code plugin that swaps in smarter file and search tools, and reported 54% lower cost, 68% fewer turns, and faster completion on the same Opus 4.7 task. The benchmark is community-run, but it includes install steps and repeatable commands.
TL;DR
- hasantoxr's benchmark post said WOZCODE cut one Claude Code run from $14.38 to $6.61, 236 turns to 76, and 22 minutes 44 seconds to 15 minutes 46 seconds on the same Opus 4.7 task.
- According to hasantoxr's tool explanation and heyrimsha's follow-up, the claimed savings come from replacing Claude Code's generic file and search tools with higher-context versions that reduce round-trips.
- hasantoxr's install post framed WOZCODE as a Claude Code plugin with marketplace install commands, while the GitHub repo adds a restart step, a
woz:codestatus badge, and browser login flow. - hasantoxr's TerminalBench claim put WOZCODE at 70% on TerminalBench 2.0, and the official site now markets it as beating raw Claude Code on performance, speed, and cost.
You can check the official product site, browse the plugin repo, and compare the claim against the live Terminal-Bench 2.0 leaderboard. The repo also exposes a few extra details the tweets skipped, including subagents, a /woz-savings command, and optional Conductor support.
Benchmark deltas
The core claim is unusually specific: one stock Claude Code run versus one WOZCODE run, same prompt, same Claude Opus 4.7 model. In that comparison, hasantoxr's benchmark post reported four deltas:
- Cost: $14.38 → $6.61, 54% lower
- Turns: 236 → 76, 68% fewer
- Duration: 22m 44s → 15m 46s, 31% faster
- Output tokens: 102k → 77.4k, 24% lower
heyrimsha's post repeated the same numbers a few minutes later, which at least suggests the benchmark format was easy to copy, even if it is still community-run rather than a formal eval.
Tool swaps
The argument behind the savings is straightforward. Each grep, file read, or shell call becomes another model round-trip, and both hasantoxr's explanation and heyrimsha's follow-up said Claude Code's default tools often return too little context, which forces the agent to ask again.
The GitHub repo describes WOZCODE the same way: it replaces Claude Code's built-in file tools with optimized alternatives. The official site turns that into a bigger marketing claim, saying the plugin is 30 to 40 percent faster on most tasks, up to 10x faster on database tasks, and cheaper on token usage.
Install path
The install path is short enough to fit in one post. Inside Claude Code, hasantoxr's install post says to run:
/plugin marketplace add WithWoz/wozcode-plugin/plugin install woz@wozcode-marketplace
The repo README adds the parts that matter once the plugin is installed:
- restart Claude Code with
claude - look for a
woz:codebadge in the input area - log in with
/woz-login - use
/woz-savingsto see estimated roundtrip, time, token, and cost savings
hasantoxr's install post also claimed the free tier includes $100 per month in Claude Code savings and said code stays on the local machine plus Anthropic's API, with no proxy or intercept.
Hidden extras
The tweets mostly sold WOZCODE as a cheaper tool layer. The repo README shows a broader product surface:
woz:codeis the main coding agentwoz:explorehandles fast read-only codebase explorationwoz:planhandles architecture and implementation planning/woz-settingscan change attribution, status line, and spinner behavior/reload-pluginsupdates the current session without a full reinstallwozcode conductorwires the plugin into Conductor
That last part is new information relative to the tweet thread. WOZCODE is being pitched as a drop-in Claude Code plugin, but the repo is already describing subagents, settings, and external tool integration that make it look more like a lightweight agent layer riding inside Claude Code.