测试显示"免费"AI 工具 Argent 比 Buoy 慢 2.2 倍、AI 账单高 2.2 倍
'free' AI tool is costing you money
Buoy 团队用 Claude Sonnet 5.5 在 Claude Code 中测试了 Argent 0.27.0 与 Buoy MCP 7.0.63：Argent 耗时是 Buoy 的 2.2 倍，AI 账单也是 2.2 倍，14 项任务全部慢于 Buoy。
Test results
In our test, the free tool took 2.2× longer than Buoy and ran up 2.2× the AI bill.
Updated October 9, 2026 · by the Buoy team
faster than Argent
the AI bill of Argent
faster than no tools
The race
Claude Sonnet 5.5 did 14 jobs in our test app, 3 times each. It went once with no tools, once with Argent and once with Buoy MCP.
all 14 jobs, min:sec
Buoy
Argent6:26
No tools17:44
every run
Buoy
Argent$5.71
No tools$5.92+
Runs passedBuoy 41 of 42Argent 40 of 41No tools 31 of 42
No tools cost at least $5.92, because 10 runs ran out of time. Buoy used its Pro tools. We tuned Buoy on these jobs.
The film
The whole race as a two minute carnival horror film.
Why Buoy is faster
Other tools
Read the screen. Tap. Read the screen again.
They work from the outside, so the AI guesses from what it can see.
Buoy
Read the app's data and web calls. Right away.
So the AI makes fewer guesses and burns fewer tokens.
What's your hour worth?
A senior mobile dev makes about $120,000 to $160,000 a year. A lead can make up to $200,000. Here is what this test adds up to, next to Argent.
Small team
6 devs · 2 web, 2 mobile, 2 backend
a year · 10 hours of AI waiting back a month
Mid-size team
25 devs · a few teams
a year · 42 hours of AI waiting back a month
Big company
75 devs · 50 to 100 devs
a year · 125 hours of AI waiting back a month
Estimates, not a promise. Each dev's AI does 20 app jobs a day, 20 work days a month. A dev costs about $72 an hour (a $150,000 salary). Each job saves the average from this test: about 15 s and 8¢ vs Argent. Next to no tools, the small team gets back about 42 hours and $3,238 a month.
Buoy Pro
Those numbers only count your AI agent. Pro gives your whole team the same tools: in the app, on Buoy Desktop and through your AI. We can't put a price on that yet.
Every person who stops waiting on a dev adds up. That is the multiplier.
Free: your AI reads the screen, taps, swipes, types and reads logs. Pro: it sees inside your app.
Every number
We make Buoy. So here is how we tested, every job and where Buoy lost.
This is the middle of 3 runs. The fastest time for each job is in green. About 150 s means at least 2 of the 3 runs ran out of time.
| Job | No tools | Argent | Buoy MCP |
|---|---|---|---|
| Read one number off the screen | 5.5 s | 46.4 s | 7.1 s |
| Add 3 burgers to the cart | 150.3 s | 20.8 s | 15.2 s |
| Find why checkout gets declined | 112.2 s | 28.8 s | 20.6 s |
| Set the crowns to 1,000,000 | 81.5 s | 40.4 s | 16.9 s |
| Read the last web call | 31.6 s | 16.0 s | 6.8 s |
| Save a test and run it twice | 150.3 s | 103.8 s | 44.8 s |
| Tap a tab | 30.7 s | 11.7 s | 5.6 s |
| Type a name | 75.7 s | 29.1 s | 9.4 s |
| Scroll to the end | 71.8 s | 14.7 s | 11.3 s |
| Pinch and turn the map | 39.2 s | 14.2 s | 7.9 s |
| Turn the phone sideways | 150.2 s | 6.0 s | 4.3 s |
| Record 3 taps | 132.8 s | 21.2 s | 12.9 s |
| Say what's under a finger | 15.0 s | 7.8 s | 6.2 s |
| Spot what changed on the screen | 16.9 s | 24.8 s | 6.8 s |
Buoy was faster than Argent on all 14 jobs.
Easy jobs don't need a tool. To read one number off the screen, no tools took a screenshot and read it in 5.5 s. Buoy took 7.1 s, and Argent took 46.4 s. No tools was also faster than Argent at spotting what changed on the screen: 16.9 s vs 24.8 s. Buoy took 6.8 s.
- We used Claude Sonnet 5.5 in Claude Code. It worked on our own food ordering test app on iOS Simulators. Buoy is in that app.
- We used Argent 0.27.0 and Buoy MCP 7.0.63. Every AI also had a shell.
- Each job ran 3 times for each tool, with a 150 second limit. A script checked the app after each run.
- Times are the middle of the 3 runs. The totals add those up.
With no tools, 10 runs ran out of time. They logged $0, because they were stopped before the bill came in. But they still used tokens. So we priced their tokens from the run logs, at Sonnet prices. That came to $1.68. The same math on the runs that finished matched the real bill within 0.2%. So no tools cost at least $5.92.
- We planned 15 jobs. We dropped one, because an old code change in the app had already done it for all three.
- One Argent run was not scored, because our checker failed. So Argent has 41 runs, not 42.
- We tuned Buoy on these jobs. Keep that in mind.
- The race used Buoy's free tools and its Pro tools. We have not timed Free alone.
- The team numbers are a guess built from this test. They are not measurements.
- Argent is free and open source. Argent is a trademark of its owner.
What is free in Buoy MCP now?
Reading the screen, taps, swipes, typing, steps, screenshots, turning the phone, logs and the list of web calls. Pro tools see inside your app, like its data and web call bodies.
Can I use Buoy and Argent together?
Yes. In our tests, Argent ran in an app with Buoy installed.
Can I see the raw runs?
We kept the log of every run, with the tool calls, times and costs. Contact us if you want to check a number.
related
sources
Capability claims on this page come from each vendor's own documentation, read on the date shown. We did not install and run every tool listed.
- Argent repository and README (Software Mansion) · checked October 9, 2026
- Argent tools reference · checked October 9, 2026
来源：Hacker News · buoy.gg