TRACKERLicense manager4 projects · 1 agent · 2 flows
Anthropic · tested in Taskade Genesis
How Claude builds in Taskade Genesis
Claude is not in the Taskade model menu. It runs on your own Claude API key, in automations and in custom agents on Pro and up.
Last test
TSK-1 scorecard for Claude
TSK Score
Task completion across real apps built in Taskade Genesis
TSK Score 75 out of 100
- Interfaceready to share
- Strong
- Taskcompletes your task
- Strong
- Memorykeeps your data
- Emerging
- Adapthandles follow-up edits
- Leading
Best for
Polished apps that keep improving
Claude produces the most polished finished apps we test and handles follow-up changes especially well.
MoreLess
Its main risk is losing details in a long request. Opus 5.5, the newest, shared the top result of our September phone-form test.
Where does Claude rank?
TSK Score out of 100.
| Date | Claude Opus 5.5 | Claude Sonnet 5 | Claude Sonnet 4.5 | Claude Opus 5 | Claude Haiku 4.5 |
|---|---|---|---|---|---|
| 0 | 1, tested this day | 0 | 1, tested this day | 0 | |
| 0 | 1 | 0 | 1 | 1, tested this day | |
| 0 | 1 | 0 | 2, tested this day | 1 | |
| 0 | 2, tested this day | 0 | 2 | 1 | |
| 0 | 3, tested this day | 0 | 2 | 1 | |
| 0 | 4, tested this day | 0 | 3, tested this day | 2, tested this day | |
| 0 | 5, tested this day | 0 | 4, tested this day | 2 | |
| 1, tested this day | 5 | 0 | 4 | 2 | |
| 2, tested this day | 6, tested this day | 0 | 5, tested this day | 2 | |
| 2 | 7, tested this day | 1, tested this day | 5 | 3, tested this day | |
| 2 | 8, tested this day | 1 | 5 | 3 | |
| 2 | 9, tested this day | 1 | 5 | 3 |
Which Claude versions did we test?
Open a version to read its findings.
Claude Opus 5.5Tested Sep 20264 findings
The newest Claude we test. On a family intake form it saved every answer and both photos, turned away an oversized photo before the upload, and was the only build of its test to open in the warm cream look the request asked for.
Sep 22, 2026
Asked where the weekly recap should go and who would use the tracker before it built, then built three pages and a coaching automation that wrote a note into a logged match.
Sep 30, 2026
Built the family intake form so it saved every answer and both photos, and it turned away a 12 MB photo before the upload started.
Sep 30, 2026
The only build of its test that opened in the light cream look the request asked for.
Sep 30, 2026
Added the new teacher question and put the change live. Its class assistant answered questions about the class but could not count the families.
Claude Sonnet 5Tested Oct 202617 findings
Our most carefully finished result in August. In September, one dashboard could not read its own sample data and a longer build stopped before the app was written. In October we had it build each app three times: none of its family forms saved the time each form was sent, which the request asked for, and every reading list, family form and follow-up question saved.
Jul 30, 2026
Opened its own app and checked the assistant's answers: the only model to verify its work this way.
Aug 3, 2026
Most carefully finished result, with eight minor items left.
Aug 3, 2026
Built a contacts database instead of the requested sign-up form after two stalls.
Aug 6, 2026
Four attempts stalled before the sign-up form was completed.
Aug 25, 2026
Stopped after asking its questions on the client sign-up form, and wrote the plan instead of the app.
Sep 19, 2026
Typed an 81-row scoring guide and a 36-field sign-up table, then stopped at an approval prompt before writing the app.
Sep 19, 2026
Built a three-page match tracker whose dashboard reported zero matches over the eight matches it had entered itself.
Sep 30, 2026
Built the family intake form so it saved every answer and both photos, then added the new teacher question and put the change live.
Oct 8, 2026
Built the family intake form twice. Both builds saved every answer and both photos and took the new teacher question, and both showed sample families on the teacher page with no label.
Oct 8, 2026
Built the Dota 2 match tracker twice, each with a win-rate trend on the dashboard, and both saved a logged match. One ran wider than a phone screen. In the other the change added a second heroes page beside the one it had already built.
Oct 18, 2026
Built a reading list that saved a book added from the app, and each status filter showed the right books.
Oct 18, 2026
Built the family intake form so it saved every answer and both photos, and its teacher page showed each answer choice in plain words. Each saved family had a blank name in the project list.
Oct 18, 2026
Added the new teacher question, and the answer saved. Its closing summary described the teacher page as private-feeling, but anyone with the link could open it.
Oct 18, 2026
Added an Our Families page with sibling and family-activity fields to an intake app that another model had built. It locked the page with a class password written into the code of the app, and its closing summary said that this was not real security.
Oct 20, 2026
Built the reading list three times. Every build saved a book added from the app with its author and status, and each status filter showed the right books. One build could not change a book's status from the page, and one showed its Reading label almost invisibly in the dark look.
Oct 20, 2026
Built the family intake form three times. Every build saved each answer and both photos, and the new teacher question saved and showed on the teacher page each time. None saved the date and time of each submission that the request asked for, and one teacher page showed raw answer codes in place of plain words.
Oct 20, 2026
Added the Our Families page to a copy of another model's intake app three times. Each build put the page behind the app sign-in, with no password in the app. Two showed every family detail the request named, but one of them read a new sharing question and ignored the photo permission the form already had. The third left out some family details.
Claude Sonnet 4.5Tested Oct 20262 findings
An older Claude, tested in October. Its match tracker rejected every logged match and read every sample match as a loss, while its family intake form saved every answer and both photos.
Oct 8, 2026
Built the family intake form so it saved every answer and both photos, in a plain look with little of the warm pastel the request asked for. Its open teacher page listed parent phone numbers.
Oct 8, 2026
Built a Dota 2 match tracker that rejected every logged match and showed every sample match as a loss, so the dashboard read a 0% win rate.
Claude Opus 5Tested Sep 20267 findings
The most complete client sign-up build of September: five pages, every question in the original wording, and a scoring automation that ran and saved its verdict.
Jul 30, 2026
Set the design high-water mark with a polished, consistent interface.
Aug 1, 2026
Created the fullest workspace, with an assistant that understood its contents.
Aug 25, 2026
The most complete client sign-up build of the test: a five-page form, every answer saved, and the scoring automation graded the submission on its own.
Aug 25, 2026
Best-executed match tracker of the test, without a single failed build step.
Sep 19, 2026
Finished the client sign-up form with all 32 questions in the original wording, 31 word for word, five pages, and a scoring automation that saved its result.
Sep 19, 2026
Decided two open points in the brief and named both decisions in its closing summary.
Sep 30, 2026
Built the family intake form so it saved every answer and both photos, and its email automation ran three times. It got a lower result than Claude Opus 5.5 on the same request.
Claude Haiku 4.5Tested Oct 20264 findings
Caught and repaired a problem that would have left the app blank.
Jul 31, 2026
Found the problem while building, repaired it, and then finished: the only model in the test to recover on its own.
Aug 25, 2026
Fastest build of the test, but it left out the weekly recap automation the brief asked for.
Oct 8, 2026
Built the family intake form in under 4 minutes and took the new teacher question, but the family photo replaced the child photo on every submission.
Oct 8, 2026
Built a Dota 2 match tracker that saved logged matches and added a heroes page when asked, but its top-heroes chart showed no data.
Available in Taskade, not tested yet: Claude Sonnet 5.5 (on your own key).
Model facts
The dates, limits and features of each tested model, as models.dev lists them. They describe the model, not how Taskade runs it.
| Model | Released | Knowledge cutoff | Context window in tokens | Max output in tokens | Input | Reasoning | Tool calling | Open weights |
|---|---|---|---|---|---|---|---|---|
| Claude Opus 5.5 | 1M tokens | 128K tokens | Text, Images, PDF | Yes | Yes | No | ||
| Claude Sonnet 5 | 1M tokens | 128K tokens | Text, Images, PDF | Yes | Yes | No | ||
| Claude Sonnet 4.5 | 200K tokens | 64K tokens | Text, Images, PDF | Yes | Yes | No | ||
| Claude Opus 5 | 1M tokens | 128K tokens | Text, Images, PDF | Yes | Yes | No | ||
| Claude Haiku 4.5 | 200K tokens | 64K tokens | Text, Images, PDF | Yes | Yes | No |
Published specs from the model maker, via models.dev (MIT).
Published scores
The benchmark scores that labs and test sites published for each Claude model, with a link to each source. TSK-1 tests how a model builds apps. These scores test other skills.
How did Claude do on the same request?
The family intake form, three builds by one model
All models →The request
Parents fill in a form on their phone with two to four photos, the teacher sees every family on one page, and then we ask for one more question on the form.
Claude Sonnet 53 builds (same result)Works, with gaps
All three saved every answer and both photos, and the new teacher question saved each time. None saved the date and time of each submission that the request asked for, and one teacher page showed raw answer codes.
The reading list, three builds by one model
All models →The request
A simple reading list: books with a title, an author and a status (to read, reading or done), on one page that lists them and filters them by status.
Claude Sonnet 53 builds (best works, worst has gaps)Works
All three saved a book added from the app, and each status filter showed the right books. One could not change a book's status from the page, and one showed its Reading label almost invisibly in the dark look.
The family intake form, three more builds
All models →The request
Parents fill in a form on their phone with two to four photos, the teacher sees every family on one page, and then we ask for one more question on the form.
Claude Sonnet 51 buildWorks
Saved every answer and both photos, and its teacher page showed each answer choice in plain words. The new teacher question saved.
A reading list app
All models →The request
A simple reading list: books with a title, an author and a status (to read, reading or done), on one page that lists them and filters them by status.
Claude Sonnet 51 buildWorks
A book added from the app saved, and each status filter showed the right books.
The family intake form, more models
All models →The request
Parents fill in a form on their phone with two to four photos, the teacher sees every family on one page, and then we ask for one more question on the form.
Claude Sonnet 52 builds (same result)Works
Both builds saved every answer and both photos, and the new teacher question landed. Sample families showed on the teacher page with no label.
Claude Haiku 4.51 buildWorks, with gaps
Saved every answer, but the family photo replaced the child photo on every submission.
Claude Sonnet 4.51 buildWorks, with gaps
Saved every answer and both photos in a plain look, and its open teacher page listed parent phone numbers.
A Dota 2 match tracker with a heroes page
All models →The request
One player logs each match with hero, result, kills, deaths, assists, duration and notes, a dashboard shows win rate and streaks in a premium esports look, and then we ask for a heroes page.
Claude Sonnet 52 builds (same result)Works
Both builds put a win-rate trend on the dashboard and saved a logged match. One ran wider than a phone screen, and in the other the change added a second heroes page.
Claude Haiku 4.51 buildWorks, with gaps
Saved logged matches and added the heroes page, but its top-heroes chart showed no data.
Claude Sonnet 4.51 buildWorks, with gaps
Rejected every logged match and showed every sample match as a loss, so the win rate read 0%.
A family intake form on a phone
All models →The request
Parents fill in a form on their phone with two to four photos, the teacher sees every family on one page, and then we ask for one more question on the form.
Claude Opus 5.51 buildWorks
Saved every answer and both photos, turned away an oversized photo, and opened in the cream look the request asked for. The follow-up change landed.
Claude Opus 51 buildWorks
Saved every answer and both photos, and its email automation ran. The follow-up change landed.
Claude Sonnet 51 buildWorks
Saved every answer and both photos. The follow-up change landed.
A match tracker
All models →The request
One player logs each game with hero, result and notes, and a dashboard shows win rate and streaks.
Claude Opus 5.51 buildWorks
Asked two questions first, then built three pages and a coaching automation.
Most results come from one build. Where a model built the app more than once, we show its best and worst build. Read the results as a guide, not a final ranking.
Preview runs
Early runs from our newest tests that the ranking does not count yet. Each one used different model settings than Taskade uses, and some were design tests run by hand. Each run says whether it finished an app and what it built. They get no points and no rank, so read them as notes, not results.
12 preview runs, newest Oct 23, 2026
Claude Opus 4.84 preview runs, newest Oct 23, 2026Preview, not ranked
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: the client sign-up form with scoring.
A generic dashboard with five pages, editable score cutoffs and four labeled sample entries.
Its automation only sent an email alert, so staff scored each answer by hand. A restart in our test setup interrupted the run before its first question, so it set its own score cutoffs and said so.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: a Dota 2 match tracker from a one- or two-word request.
Three pages, eight sample matches and a public coach.
It picked a dark look that the request did not ask for, and its page ran wider than a phone screen.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: the client sign-up form with scoring.
A form with generic feature cards and a stat row, and a scoring automation that ran once.
It changed one question to allow more than one answer and said so. Its home page said submissions were private while the staff dashboard was open to anyone, and its closing summary did not mention it.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: the family intake form.
A calm cream form in one narrow column with pencil doodles, the only build of its test in the calm one-column look.
The teacher page sat behind the app sign-in. It checked a real submission before it finished.
Claude Opus 4.65 preview runs, newest Oct 23, 2026Preview, not ranked
· Preview run, not ranked: tested with different model settings than Taskade uses.
Finished an app. Asked for: a clinic patient intake form.
A plain centered form with no header, and a review page.
Its review page was open to anyone, and it offered a staff sign-in only as a suggestion.
· Preview run, not ranked: tested with different model settings than Taskade uses.
Ran out of test time. Asked for: the client sign-up form with scoring.
A plain form under a white top bar.
The test time ran out while it checked the form, before its closing summary.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: a status page from a one-word request.
A status page with a status banner, active incidents, grouped services, a history page and an admin page.
The admin page was open to anyone with the link, and its closing summary said so. In the light look the status text was hard to read.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: the client sign-up form with scoring.
A dark first view with a glowing accent, a plain light view, and AI scoring.
It chose the score cutoffs without asking and said so. On a phone the dashboard dropped the applicant name column, and a chart from its plan never appeared.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: the client sign-up form with scoring.
A generic admin dashboard with stat cards and charts, and a clean eight-section form that kept each question word for word.
Its saved fields shortened the long questions, which the request said not to do, and its three sample scores came from the model, not from the scoring automation.
Claude Sonnet 52 preview runs, newest Oct 23, 2026Preview, not ranked
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: a status page from a one-word request.
A cool slate and cyan operations dashboard with six pages, clear in the light and the dark look.
One section crashed for about a minute, and it found and fixed the crash itself. Every incident date showed one day early, and its closing summary called the app fully checked.
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: the client sign-up form with scoring.
A calm six-step form with no header, a staff page behind an admin sign-in, and a scoring automation that ran once on a test entry.
After it summarized its own chat, it thanked the user for pasting the questions again, though no new message had been sent.
Claude Sonnet 4.61 preview run, newest Oct 23, 2026Preview, not ranked
· Preview run, not ranked: a design test run by hand, with different model settings than Taskade uses.
Finished an app. Asked for: a Dota 2 match tracker from a one- or two-word request.
A generic tracker: a small title, four stat tiles and a list.
Its red accent sat close to the color for a loss, it picked a dark look that the request did not ask for, and its closing summary said that a digest emailed the workspace when none did.
How we grade these apps: read the full method.
App kits you can open today
Live apps from the official Taskade account, not builds by Claude.
License manager Lead pipeline tracker CRMLead pipeline tracker4 projects · 2 agents · 3 flowsInvoice tracker TOOLInvoice tracker1 project · 1 agent · 2 flowsClient portal PORTALClient portal3 projects · 1 agent · 4 flowsInvestor metrics dashboard DASHBOARDInvestor metrics dashboard5 projects · 2 agents · 4 flowsOrder and stock tracker TRACKEROrder and stock tracker2 projects · 1 agent · 3 flows
How does Claude compare?
See all 11 comparisonsShow fewer
Other models in TSK-1
FAQ
What is Claude's TSK Score?
Claude scores 75 out of 100 on TSK-1: Interface Strong, Task Strong, Memory Emerging, Adapt Leading. Last test Oct 20, 2026.
What is Claude best at in Taskade Genesis?
Claude is strongest when finish and follow-up changes matter. Sonnet produced our most carefully finished result, while Opus set the visual quality high-water mark. With long requests, review the finished app to make sure every detail carried through.
Which Claude model is best for app building?
Claude Opus 5.5 shared the top result of our September phone-form test and opened in the look the request asked for. Claude Opus 5 produced the more complete September sign-up app, with five pages and a scoring automation that ran. Claude Sonnet 5 remains strong at careful finish and follow-up changes, but review its dashboards against the saved data.
Why did Claude fail a test?
In one August test it built a contacts database instead of the client sign-up form, after two three-minute stalls lost it the thread. In another, four stalled runs in a row meant the client sign-up form never got built at all. We publish the tests that go badly alongside the ones that go well, and we open every app and read it back against the brief. That is the point of the benchmark.
Can I use Claude in Taskade?
Yes, with your own Claude API key: use it in the Anthropic Claude automation step, or connect it to your custom agents on Pro and up. Claude is not in the Taskade model menu.
Where can I compare Claude with other models, or try it myself?
Side-by-side model pages live at /compare. Claude is not in the model menu, so /create starts a build with Auto or another available model. This page stays on what Claude produced in TSK-1 tests.