Papers › WebLINX: Real-World Website Navigation with Multi-Turn Dialogue

WebLINX: Real-World Website Navigation with Multi-Turn Dialogue

8 Feb 2024arXiv:2402.05930archive 2025-07-28

Xing Han Lù, Zdeněk Kasner, Siva Reddy

We propose the problem of conversational web navigation, where a digital agent controls a web browser and follows user instructions to solve real-world tasks in a multi-turn dialogue fashion. To support this problem, we introduce WEBLINX - a large-scale benchmark of 100K interactions across 2300 expert demonstrations of conversational web navigation. Our benchmark covers a broad range of patterns on over 150 real-world websites and can be used to train and evaluate agents in diverse scenarios. Due to the magnitude of information present, Large Language Models (LLMs) cannot process entire web pages in real-time. To solve this bottleneck, we design a retrieval-inspired model that efficiently prunes HTML pages by ranking relevant elements. We use the selected elements, along with screenshots and action history, to assess a variety of models for their ability to replicate human behavior when navigating the web. Our experiments span from small text-only to proprietary multimodal LLMs. We find that smaller finetuned decoders surpass the best zero-shot LLMs (including GPT-4V), but also larger finetuned multimodal models which were explicitly pretrained on screenshots. However, all finetuned models struggle to generalize to unseen websites. Our findings highlight the need for large multimodal models that can generalize to novel settings. Our code, data and models are available for research: https://mcgill-nlp.github.io/weblinx

PaperPDFCodeCode Syntology ran

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 27 code samples harvested from 2 repositories linked to this paper: 21 ran (of which 0 constructed an object rather than computing a result; 12 with no instrument failure: 2 honoured, 1 violated, 9 with no contract checked; 9 where Syntology's instrument failed) · 6 unverified. Where Syntology's instrument failed: 8 where our contract draft was wrong; 1 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: official repository: 27 samples from 2 repositories, 21 ran (12 with no instrument failure; 9 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Levels written between 2026-09-22 and 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

McGill-NLP/weblinx officialmentioned on GitHubpytorchApache-2.0Syntology: 23 samples, 19 ran (8 instrument-failed) · 4 unverified report
McGill-NLP/webllama officialpytorchSyntology: 4 samples, 2 ran (1 instrument-failed) · 2 unverified report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

27 samples harvested: 21 ran (of which 0 constructed an object rather than computing a result; 12 with no instrument failure: 2 honoured, 1 violated, 9 with no contract checked; 9 where Syntology's instrument failed) · 6 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

2ran · honoured contract
1ran · violated contract
8ran · our draft was wrong
1ran; the contract check could not run
9ran
6unverified

Licence: 0 of the 27 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from 2 repositories linked to this paper, official or community; the list below is grouped by repository, each group headed by its name and what it is to the paper. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity. A census tag appears when Syntology's census ran implementations with that name on one shared input per bucket: it gives how many were compared, from how many papers, and how many distinct outputs they produced, and names this sample's bucket when this sample was one of them.

McGill-NLP/weblinx official repository 23 samples, 19 ran (8 instrument-failed) · 4 unverified
cast_to_float weblinx/processing/outputs.py ran · honoured contract fingerprinted Apache-2.0 (permissive) · fa80bfc9b9704f8c · report fixture: signature value:float/0/float · returned float · level recorded 2026-09-24
compose_formatters weblinx/utils/format.py ran · our draft was wrong Apache-2.0 (permissive) · 1ec7b369dd522796 · report fixture: signature formatters:list/1/callable · returned function · level recorded 2026-09-24
filter_bboxes weblinx/utils/html.py ran Apache-2.0 (permissive) · 4fcb1951b117a6e6 · report fixture: signature bboxes:dict/0/?, min_height:int/0/int, min_width:int/0/int, viewport_height:int/0/int, viewport_width:int/0/int · returned dict · level recorded 2026-09-24
format_action_arg_value weblinx/processing/outputs.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · bc51f017163fe65f · report fixture: signature arg_value:str/0/str · returned str · level recorded 2026-09-24
format_arg_item weblinx/utils/format.py ran Apache-2.0 (permissive) · 4d03c832e78532b4 · report fixture: signature turn:Turn/0/?, name:str/0/str, strip:bool/0/bool, max_length:int/0/int, return_as:str/0/str · returned dict · level recorded 2026-09-24
format_prev_turns weblinx/processing/prompt.py ran Apache-2.0 (permissive) · a899f2c938e9474d · report fixture: signature replay:Replay/0/?, turn:Turn/0/?, format_intent:callable/0/?, turn_sep:str/0/str, num_prev_turns:int/0/int, return_as:callable/0/? · returned str · level recorded 2026-09-24
get_bracket_length weblinx/processing/truncation.py ran Apache-2.0 (permissive) · c6518874884b35e2 · report fixture: signature elem_open_bracket:str/0/str, elem_close_bracket:str/0/str, tokenizer:StandInTokenizer/0/? · returned int · level recorded 2026-09-24
get_descendants in 5 papers weblinx/processing/dom.py ran · our draft was wrong Apache-2.0 (permissive) · 44d3924c80fe3091 · report fixture: signature node:TreeNode/0/?, max_depth:int/0/int, current_depth:int/0/int · returned list · level recorded 2026-09-24
get_env_dict weblinx/utils/envs.py ran · our draft was wrong Apache-2.0 (permissive) · abbcbd290c538f35 · report fixture: signature filepath:str/0/str · returned dict · level recorded 2026-09-24
get_list_from_records_by_key weblinx/utils/recs.py ran · our draft was wrong Apache-2.0 (permissive) · b070440172fb7a66 · report fixture: signature records:list/1/dict, key:str/0/str · returned list · level recorded 2026-09-24
get_speaker weblinx/processing/prompt.py ran Apache-2.0 (permissive) · 576db6c2c80cba48 · report fixture: signature utterance:str/0/str, instructor_name:str/0/str, navigator_name:str/0/str, default_name:NoneType/0/? · returned str · level recorded 2026-09-24
get_truncation_offsets weblinx/processing/truncation.py ran Apache-2.0 (permissive) · e8d383f64048fb06 · report fixture: signature tokens:dict/0/?, length:int/0/int, num_tokens_to_remove:int/0/int · returned tuple · level recorded 2026-09-24
group_record_to_dict weblinx/utils/recs.py ran Apache-2.0 (permissive) · 176890ae96f63695 · report fixture: signature records:list/1/dict, keys:list/1/str, remove_keys:bool/0/bool, copy:bool/0/bool · returned dict · level recorded 2026-09-24
has_elem_in_viewport weblinx/utils/html.py ran Apache-2.0 (permissive) · 1fec2e813060a926 · report fixture: signature turn:Turn/0/?, attrib_dict:dict/0/?, key:str/0/str, min_height:int/0/int, min_width:int/0/int, verbose:bool/0/bool · returned bool · level recorded 2026-09-24
identity in 174 papers
census: 11 implementations from 39 papers, 3 distinct outputs across 2 buckets, one shared input per bucket
weblinx/processing/prompt.py ran · honoured contract fingerprinted Apache-2.0 (permissive) · 6b063e5a47833dd4 · report fixture: signature x:Tensor/2/float · returned Tensor · level recorded 2026-09-24
is_list_monotonically_increasing weblinx/utils/recs.py ran; the contract check could not run (RAISES) fingerprinted Apache-2.0 (permissive) · 3ae8877f0adf8db3 · report fixture: signature lst:list/1/int · returned bool · level recorded 2026-09-24
list_all_non_alphanum_chars weblinx/processing/outputs.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · dcda9d27514120e0 · report fixture: signature other_chars_allowed:list/1/str · returned list · level recorded 2026-09-24
open_html_with_encodings weblinx/utils/html.py ran Apache-2.0 (permissive) · 63c7e3d03045c0db · report fixture: signature path:str/0/str, encodings:list/1/str, raise_error:bool/0/bool · returned str · level recorded 2026-09-24
shorten in 5 papers weblinx/utils/format.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · d20962095e5b96d0 · report fixture: signature text:str/0/str, max_length:int/0/int · returned str · level recorded 2026-09-24
get_tree_repr_simple weblinx/processing/dom.py unverified Apache-2.0 (permissive) · af79280883191a10 · report
prune_tree weblinx/processing/dom.py unverified Apache-2.0 (permissive) · 7c1d79a2be5ce989 · report
reduce_list_of_lengths weblinx/processing/truncation.py unverified Apache-2.0 (permissive) · 0ac8898c0f3f43ec · report
resolve_cache_path weblinx/utils/hydra.py unverified Apache-2.0 (permissive) · 736f8a0adf254065 · report
McGill-NLP/webllama official repository 4 samples, 2 ran (1 instrument-failed) · 2 unverified
remap_bboxes examples/browsergym/agent.py ran · our draft was wrong MIT (permissive) · 84bd76bf303ce78b · report fixture: signature bboxes:dict/0/?, attrs_map:dict/0/? · returned dict · level recorded 2026-09-22
remove_latex app/Results.py ran · violated contract fingerprinted MIT (permissive) · d31d4a5938635d95 · report fixture: signature x:str/0/str · returned str · level recorded 2026-09-22
build_cond_df app/Results.py unverified MIT (permissive) · 168f0e5f647cc7c9 · report
load_and_clean_df app/Results.py unverified MIT (permissive) · 78b8c5d1e64ad4a3 · report

Tasks

Conversational Web NavigationText GenerationVision and Language Navigation

Datasets

Introduced by this paper, per the archive.

WebLINX

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Conversational Web Navigation WebLINX Llama-2-13B Element (IoU) 22.82 #1 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-13B Intent Match 81.91 #1 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-13B Overall score 25.21 #1 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-13B Text (F1) 26.60 #1 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-2.7B Element (IoU) 22.60 #2 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-2.7B Intent Match 84.00 #2 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-2.7B Overall score 25.02 #2 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-2.7B Text (F1) 27.17 #2 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-7B Element (IoU) 22.26 #3 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-7B Intent Match 82.64 #3 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-7B Overall score 24.57 #3 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Llama-2-7B Text (F1) 26.50 #3 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-3B Element (IoU) 20.31 #4 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-3B Intent Match 81.14 #4 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-3B Overall score 23.77 #4 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-3B Text (F1) 25.75 #4 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-1.3B Element (IoU) 20.54 #5 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-1.3B Intent Match 83.32 #5 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-1.3B Overall score 23.73 #5 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX S-LLaMA-1.3B Text (F1) 25.85 #5 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5F Element (IoU) 18.64 #6 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5F Intent Match 77.56 #6 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5F Overall score 21.22 #6 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5F Text (F1) 22.39 #6 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-3B Element (IoU) 16.50 #7 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-3B Intent Match 79.89 #7 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-3B Overall score 20.94 #7 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-3B Text (F1) 23.16 #7 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Fuyu-8B Element (IoU) 15.70 #8 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Fuyu-8B Intent Match 80.07 #8 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Fuyu-8B Overall score 19.97 #8 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Fuyu-8B Text (F1) 22.30 #8 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-780M Element (IoU) 15.36 #9 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-780M Intent Match 80.02 #9 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-780M Overall score 17.27 #9 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-780M Text (F1) 14.05 #9 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-1.3B Element (IoU) 8.28 #10 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-1.3B Intent Match 81.80 #10 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-1.3B Overall score 16.88 #10 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-1.3B Text (F1) 25.21 #10 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-780M Element (IoU) 13.39 #11 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-780M Intent Match 75.87 #11 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-780M Overall score 15.13 #11 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-780M Text (F1) 13.58 #11 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-250M Element (IoU) 14.86 #12 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-250M Intent Match 79.69 #12 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-250M Overall score 14.99 #12 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Flan-T5-250M Text (F1) 9.21 #12 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-250M Element (IoU) 12.05 #13 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-250M Intent Match 74.25 #13 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-250M Overall score 12.63 #13 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX MindAct-250M Text (F1) 7.67 #13 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-282M Element (IoU) 6.20 #14 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-282M Intent Match 79.71 #14 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-282M Overall score 12.51 #14 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX Pix2Act-282M Text (F1) 16.40 #14 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4T (Zero-Shot) Element (IoU) 10.85 #15 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4T (Zero-Shot) Intent Match 41.66 #15 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4T (Zero-Shot) Overall score 10.72 #15 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4T (Zero-Shot) Text (F1) 6.75 #15 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4V (Zero-Shot) Element (IoU) 10.91 #16 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4V (Zero-Shot) Intent Match 42.36 #16 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4V (Zero-Shot) Overall score 10.45 #16 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-4V (Zero-Shot) Text (F1) 6.21 #16 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5T (Zero-Shot) Element (IoU) 8.62 #17 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5T (Zero-Shot) Intent Match 42.77 #17 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5T (Zero-Shot) Overall score 8.51 #17 of 17 Archive leaderboard report
Conversational Web Navigation WebLINX GPT-3.5T (Zero-Shot) Text (F1) 3.45 #17 of 17 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections