SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.07972 · 2024

OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments

arXiv · PDF · Open in the Atlas

Code that ran

10 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 6 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2404.07972.json · badge: /api/ran/2404.07972.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
xlang-ai/OSWorld canonical (classifier label)
classifier basis: census
10 ran · 0 no recorded run
FunctionStatusWhere it lives
convert_to_pyautogui_action Ran
ran; our contract draft was wrong, not the code
xlang-ai/OSWorld/gemini_action_parser.py
code served (permissive licence) · get_code("254ea70ebc95e0fd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import random

# Set seed for reproducibility
random.seed(42)

# Create a valid action dict for the convert_to_pyautogui_action function
# The function expects an 'action' parameter which is a dict with 'action_type' and 'parameters'

ARGS = {
    'action': {
        'action_type': 'click',
        'parameters': {
            'x': 500,
            'y': 300
        }
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
encode_image Ran
ran; our contract draft was wrong, not the code
xlang-ai/OSWorld/mm_agents/agent.py
code served (permissive licence) · get_code("78ee6b46cebd99ee")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64

# The function encode_image takes image_content (bytes) and base64 encodes it
# We need to provide bytes as the image_content parameter

image_content = b"This is sample image data in bytes format"

ARGS = {
    "image_content": image_content
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
encoded_img_to_pil_img Ran
ran, honoured the contract we drafted
xlang-ai/OSWorld/mm_agents/agent.py
code served (permissive licence) · get_code("4f7312030f5ef2a1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned PngImageFile. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64
from io import BytesIO
from PIL import Image
import numpy as np

# Create a small test image
img_array = np.random.randint(0, 256, (10, 10, 3), dtype=np.uint8)
pil_img = Image.fromarray(img_array)

# Encode it to base64
buffer = BytesIO()
pil_img.save(buffer, format='PNG')
img_data = base64.b64encode(buffer.getvalue()).decode('utf-8')
encoded_data_str = f'data:image/png;base64,{img_data}'

ARGS = {
    'data_str': encoded_data_str
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PngImageFile — a PngImageFile output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PngImageFile output the recorded ndim 3 and dtype family int describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_domain_from_path Ran
ran; our contract draft was wrong, not the code
xlang-ai/OSWorld/lib_results_logger.py
code served (permissive licence) · get_code("91170927d6273dd6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from pathlib import Path

ARGS = {
    "result_path": "results/discrete/image/ppo/cartpole/task_001/"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_result Ran
ran, honoured the contract we drafted
xlang-ai/OSWorld/mm_agents/maestro/osworld_run_maestro.py
code served (permissive licence) · get_code("189a37dc14c3d30a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
import shutil

# Create a temporary directory structure for testing
temp_base = tempfile.mkdtemp()

# Set up the nested directory structure that get_result expects
action_space = "discrete"
observation_type = "image"
result_dir = temp_base

target_dir = os.path.join(result_dir, action_space, observation_type)
os.makedirs(target_dir, exist_ok=True)

# Create a domain directory
domain_name = "domain1"
domain_path = os.path.join(target_dir, domain_name)
os.makedirs(domain_path, exist_ok=True)

# Create example directories with result.txt files
for example_id in ["example_0", "example_1"]:
    example_path = os.path.join(domain_path, example_id)
    os.makedirs(example_path, exist_ok=True)
    
    # Write a result.txt file with a float value
    result_file = os.path.join(example_path, "result.txt")
    with open(result_file, "w") as f:
        f.write("0.85")

# Create another domain with results
domain_name2 = "domain2"
domain_path2 = os.path.join(target_dir, domain_name2)
os.makedirs(domain_path2, exist_ok=True)

for example_id in ["example_0"]:
    example_path = os.path.join(domain_path2, example_id)
    os.makedirs(example_path, exist_ok=True)
    
    result_file = os.path.join(example_path, "result.txt")
    with open(result_file, "w") as f:
        f.write("0.95")

ARGS = {
    "action_space": "discrete",
    "observation_type": "image",
    "result_dir": temp_base,
    "total_file_json": None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_result Ran
ran, honoured the contract we drafted
xlang-ai/OSWorld/show_result.py
code served (permissive licence) · get_code("0f5d30c04a47f46a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
import shutil

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()

# Create the nested directory structure that get_result expects
# Structure: result_dir/action_space/observation_type/use_model/domain/example_id/result.txt
action_space = "pyautogui"
observation_type = "screenshot"
use_model = "gpt-4o"

base_path = os.path.join(temp_dir, action_space, observation_type, use_model)

# Create some sample domains with results
domains = ["chrome", "gimp", "libreoffice_calc", "libreoffice_impress", "libreoffice_writer", "multi_apps", "os", "thunderbird", "vlc", "vs_code"]

for domain in domains:
    domain_path = os.path.join(base_path, domain)
    os.makedirs(domain_path, exist_ok=True)
    
    # Create a few example results for each domain
    for example_id in range(1, 3):
        example_path = os.path.join(domain_path, f"example_{example_id}")
        os.makedirs(example_path, exist_ok=True)
        
        # Write result.txt with a float value (success rate, 0 or 1)
        result_value = 1.0 if example_id % 2 == 0 else 0.5
        with open(os.path.join(example_path, "result.txt"), "w") as f:
            f.write(str(result_value))

ARGS = {
    "action_space": action_space,
    "use_model": use_model,
    "observation_type": observation_type,
    "result_dir": temp_dir,
    "show_detailed_scores": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_unfinished Ran
ran on a synthesized input
xlang-ai/OSWorld/mm_agents/maestro/osworld_run_maestro.py
code served (permissive licence) · get_code("66d512885244b2c1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
import shutil

# Create a temporary directory structure for testing
temp_dir = tempfile.mkdtemp()

# Create the directory structure that get_unfinished expects
result_dir = temp_dir
action_space = "discrete"
observation_type = "image"

# Create target directory structure
target_dir = os.path.join(result_dir, action_space, observation_type)
os.makedirs(target_dir, exist_ok=True)

# Create a domain directory with some example subdirectories
domain1 = "navigation"
domain1_path = os.path.join(target_dir, domain1)
os.makedirs(domain1_path, exist_ok=True)

# Create example directories - some finished (with result.txt), some unfinished
example1_path = os.path.join(domain1_path, "example_001")
os.makedirs(example1_path, exist_ok=True)
with open(os.path.join(example1_path, "result.txt"), "w") as f:
    f.write("result")

example2_path = os.path.join(domain1_path, "example_002")
os.makedirs(example2_path, exist_ok=True)
with open(os.path.join(example2_path, "temp_file.txt"), "w") as f:
    f.write("temp")

# Create another domain
domain2 = "grasping"
domain2_path = os.path.join(target_dir, domain2)
os.makedirs(domain2_path, exist_ok=True)

example3_path = os.path.join(domain2_path, "example_003")
os.makedirs(example3_path, exist_ok=True)
with open(os.path.join(example3_path, "result.txt"), "w") as f:
    f.write("result")

# Create total_file_json with test data
total_file_json = {
    "navigation": ["example_001", "example_002", "example_003"],
    "grasping": ["example_003", "example_004", "example_005"],
    "other_domain": ["example_006"]
}

ARGS = {
    "action_space": action_space,
    "observation_type": observation_type,
    "result_dir": result_dir,
    "total_file_json": total_file_json
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
save_to_tmp_img_file Ran
ran; our contract draft was wrong, not the code
xlang-ai/OSWorld/mm_agents/agent.py
code served (permissive licence) · get_code("32fccbc780176119")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64
import os
import tempfile
from io import BytesIO
from PIL import Image
import numpy as np

# Create a small test image
img_array = np.random.randint(0, 256, (10, 10, 3), dtype=np.uint8)
img = Image.fromarray(img_array)

# Convert image to base64 string
img_buffer = BytesIO()
img.save(img_buffer, format='PNG')
img_buffer.seek(0)
img_base64 = base64.b64encode(img_buffer.read()).decode('utf-8')
data_str = f'data:image/png;base64,{img_base64}'

ARGS = {
    'data_str': data_str
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
setup_eval_logger Ran
ran; our contract draft was wrong, not the code
xlang-ai/OSWorld/lib_run_single.py
code served (permissive licence) · get_code("283989106587117f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Logger. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import logging
import os
import tempfile

# Create a temporary directory for the example_result_dir
temp_dir = tempfile.mkdtemp()

# Define the arguments
example = {
    'id': 'test_example_001'
}

example_result_dir = temp_dir

ARGS = {
    'example': example,
    'example_result_dir': example_result_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Logger — a Logger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Logger output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
setup_logger Ran
ran; our contract draft was wrong, not the code
xlang-ai/OSWorld/lib_run_single.py
code served (permissive licence) · get_code("7222d802b397cc6d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Logger. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import logging
import os
import tempfile

# Create a temporary directory for the example result
example_result_dir = tempfile.mkdtemp()

# Create a minimal example dict with an 'id' key
example = {
    'id': 'test_example_001'
}

ARGS = {
    'example': example,
    'example_result_dir': example_result_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Logger — a Logger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Logger output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 5 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 5 later) and 5 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
encode_image 4 other papers
4 later
2407.10956 (2024-07)
Spider2-V: How Far Are Multimodal Agents From Automating Data Science and Engineering Workflows?
2505.21936 (2025-05)
RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments
2510.04039 (2025-10)
GUI-Spotlight: Adaptive Iterative Focus Refinement for Enhanced GUI Visual Grounding
and 1 more not named here
encoded_img_to_pil_img 3 other papers
3 later
2407.10956 (2024-07)
Spider2-V: How Far Are Multimodal Agents From Automating Data Science and Engineering Workflows?
2505.21936 (2025-05)
RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments
2604.27776 (2026-04)
WindowsWorld: A Process-Centric Benchmark of Autonomous GUI Agents in Professional Cross-Application Environments
save_to_tmp_img_file 3 other papers
3 later
2407.10956 (2024-07)
Spider2-V: How Far Are Multimodal Agents From Automating Data Science and Engineering Workflows?
2505.21936 (2025-05)
RedTeamCUA: Realistic Adversarial Testing of Computer-Use Agents in Hybrid Web-OS Environments
2604.27776 (2026-04)
WindowsWorld: A Process-Centric Benchmark of Autonomous GUI Agents in Professional Cross-Application Environments
extract_domain_from_path 1 other paper
1 later
2606.31270 (2026-06)
Learning from Failure: Inference-Time Self-Improvement for Computer-Use Agents
setup_logger 1 other paper
1 later
2606.31270 (2026-06)
Learning from Failure: Inference-Time Self-Improvement for Computer-Use Agents

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Autonomous agents that accomplish complex computer tasks with minimal human interventions have the potential to transform human-computer interaction, significantly enhancing accessibility and productivity. However, existing benchmarks either lack an interactive environment or are limited to environments specific to certain applications or domains, failing to reflect the diverse and complex nature of real-world computer use, thereby limiting the scope of tasks and agent scalability. To address this issue, we introduce OSWorld, the first-of-its-kind scalable, real computer environment for multimodal agents, supporting task setup, execution-based evaluation, and interactive learning across various operating systems such as Ubuntu, Windows, and macOS. OSWorld can serve as a unified, integrated computer environment for assessing open-ended computer tasks that involve arbitrary applications. Building upon OSWorld, we create a benchmark of 369 computer tasks involving real web and desktop apps in open domains, OS file I/O, and workflows spanning multiple applications. Each task example is derived from real-world computer use cases and includes a detailed initial state setup configuration and a custom execution-based evaluation script for reliable, reproducible evaluation. Extensive evaluation of state-of-the-art LLM/VLM-based agents on OSWorld reveals significant deficiencies in their ability to serve as computer assistants. While humans can accomplish over 72.36% of the tasks, the best model achieves only 12.24% success, primarily struggling with GUI grounding and operational knowledge. Comprehensive analysis using OSWorld provides valuable insights for developing multimodal generalist agents that were not possible with previous benchmarks. Our code, environment, baseline models, and data are publicly available at https://os-world.github.io.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2404.07972")
get_code_for_paper("2404.07972")
have("2404.07972")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2404.07972.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2404.07972.svg)](https://syntology.ai/paper/2404.07972)

Connect an agent — have() is free.