collection of nanobrowser prompts

prompt.md

huggingface papers

prompt:

- Go to https://huggingface.co/papers and click through each of the top 3 upvoted papers.
- For each paper:
  - Record the title, URL and upvotes
  - Summarise the abstract section
- Finally, compile together a summary of all 3 papers, ranked by upvotes

model	nb ver.	result	remark
Ophiuchi-Qwen3-14B	0.1.5	100%
Devstral-small	0.1.6	100%
Falcon-H1	0.1.8	100%

prompt:

- Go to https://huggingface.co/papers and click through each of the top 3 upvoted papers.
- For each paper:
  - Record the title, URL and upvotes
  - Summarise the abstract section
  - After that go back one page
- Finally, compile together a summary of all 3 papers, ranked by upvotes

model	nb ver.	result	remark
jan-nano	0.1.7, 0.1.8	100%	need adjust prompt
Arch-Agent	0.1.7	30%	failed after 2nd paper, click wrong link and tries to login
Qwen3-coder	0.1.8	100%
Qwen3-30b-a3b	0.1.8	100%

techcrunch news

prompt:

Go to TechCrunch and extract top 10 headlines from the last 24 hours

model	nb ver.	result	remark
Ophiuchi-Qwen3-14B	0.1.5	100%
Devstral-small	0.1.6	90%	but lists 11 items instead of 10
jan-nano	0.1.7	50%	got 7 articles, but not the most recent ones, and hallucinated 3
Falcon-H1	0.1.8	90%	only got 5 headlines, not 5 extra from the latest news
Qwen3-30b-a3b	0.1.8	75%	only got 8, and loops beccause it doesn't pass the validation

prompt:

Go to https://techcrunch.com/latest/ and extract top 10 latest news from the last 24 hours on that page, scroll down when needed

model	nb ver.	result	remark
jan-nano	0.1.7	80%	got 8 articles and hallucinated 2, probably because it needs to scroll down.
Qwen3-30b-a3b	0.1.8	0%	tries to get headlines from page 2, validator catches that but planner can't.

github python repo

prompt:

Look for the trending Python repositories on GitHub with most stars

model	nb ver.	result	remark
Ophiuchi-Qwen3-14B	0.1.5	0%	cannot find the trending repo page
Devstral-small	0.1.6	0%	cannot find the trending repo page

prompt:

go to https://github.com/trending and look for the trending Python repositories with most stars, scroll down if needed.

model	nb ver.	result	remark
Ophiuchi-Qwen3-14B	0.1.5	0%	not scrolling down to get the whole page content, even when explicitly told to
Devstral-small	0.1.6	100%
jan-nano	0.1.7, 0.1.8	100%
Falcon-H1	0.1.8	0%	found 3 repo none of which is python and hallucinated stars
Qwen3-30b-a3b	0.1.8	100%

ebay shopping

prompt:

go to ebay and find the cheapest epyc milan 64-core cpu

model	nb ver.	result	remark
Ophiuchi-Qwen3-14B	0.1.5	0%	msg from planner: Planning failed: Failed to invoke with structured output: Error: Invalid boolean string
Devstral-small	0.1.8	80%	search, sort but returned 3rd cheapest with wrong seller info
jan-nano	0.1.7	0%	doesn't submit search
Falcon-H1	0.1.8	0%	does not click on submit
Qwen3-30b-a3b	0.1.8	80%	not the cheapest

prompt:

go to https://ebay.com fill in the search term "epyc milan 64-core cpu" then submit search. Find the cheapest one by sorting function of the site.

model	nb ver.	result	remark
jan-nano	0.1.7	70%	cannot sort
Falcon-H1	0.1.8	10%	does not submit search
Qwen3-30b-a3b	0.1.8	80%	not the cheapest

pubmed paper

prompt:

get the 3 most recent papers about cancer from Pubmed

model	nb ver.	result	remark
Ophiuchi-Qwen3-14B	0.1.5	50%	found 3 papers about cancer but no most recent ones
Qwen3-30b-a3b	0.1.8	100%

prompt:

get the 3 most recent papers about cancer from Pubmed use sort function of the site if needed

model	nb ver.	result	remark
Devstral-small	0.1.6	100%
Falcon-H1	0.1.8	100%

prompt:

go to https://pubmed.ncbi.nlm.nih.gov fill cancer in the search field then click submit search. Record the titles of 3 most recent papers on that page.

model	nb ver.	result	remark
jan-nano	0.1.7	100%

weather in NYC

prompt:

How's the weather in NYC?

model	nb ver.	result	remark
jan-nano	0.1.8	100%
Falcon-H1	0.1.8	100%
Qwen3-30b-a3b	0.1.8	100%