SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.07095 · ICLR · 2025

MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering

Chan Shern, Kevin Liu, James Aung, Neil Chowdhury, Oliver Jaffe, Dane Sherburn, Evan Mays, Giulio Starace, Leon Maksin, Tejal Patwardhan, Lilian Weng, Aleksander Ądry Openai

arXiv · PDF · Open in the Atlas

Code that ran

7 of 7 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; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 2 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/2410.07095.json · badge: /api/ran/2410.07095.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
openai/mle-bench canonical (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
innovatingai/automind community (classifier label)
classifier basis not recorded on the link
5 ran · 0 no recorded run
FunctionStatusWhere it lives
calculate_metrics_for_a_seed Ran
ran; our contract draft was wrong, not the code
openai/mle-bench/experiments/aggregate_grading_reports.py
pointer only (licence: NOASSERTION) · get_code("d01a33b7cb1c3b6f")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, openai/mle-bench (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

compile_prompt_to_md Ran
ran on a synthesized input
innovatingai/automind/automind/backend/utils.py
code served (permissive licence) · get_code("e06055db8ca09fe8")
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 torch
import numpy as np

# Test with various prompt types
string_prompt = "This is a simple string prompt"

list_prompt = [
    "First item",
    "Second item",
    "Third item"
]

dict_prompt = {
    "Section 1": "Content for section 1",
    "Section 2": [
        "Bullet point 1",
        "Bullet point 2"
    ],
    "Section 3": {
        "Subsection": "Nested content"
    },
    "Empty Section": None
}

ARGS = {
    "prompt": dict_prompt,
    "_header_depth": 1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — 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
exception_summary Ran
ran on a synthesized input
innovatingai/automind/automind/interpreter.py
code served (permissive licence) · get_code("740fd4892adad39f")
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 tuple. 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
import sys
import traceback

# Create a simple exception to test with
try:
    raise ValueError("Test exception message")
except ValueError as test_exception:
    captured_exception = test_exception

# Define parameters for exception_summary
working_dir = Path("/home/user/project")
exec_file_name = "test_script.py"
format_tb_ipython = False

ARGS = {
    "e": captured_exception,
    "working_dir": working_dir,
    "exec_file_name": exec_file_name,
    "format_tb_ipython": format_tb_ipython,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
IPython — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
format_time Ran
ran on a synthesized input
innovatingai/automind/automind/agent.py
code served (permissive licence) · get_code("13fd109f87ede946")
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 torch

torch.manual_seed(0)

ARGS = {
    'time_in_sec': 3661
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — 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
load_task_desc Ran
ran on a synthesized input
innovatingai/automind/automind/utils/config.py
code served (permissive licence) · get_code("611b4e88928b1693")
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 dataclasses import dataclass
from pathlib import Path
import tempfile
import os

@dataclass
class StageConfig:
    model: str
    temp: float

@dataclass
class StrategyConfig:
    strategy: str
    model: str

@dataclass
class SearchConfig:
    max_debug_depth: int
    debug_prob: float
    num_drafts: int
    hpo_prob: float
    trick_prob: float
    max_hpo_prob: float
    greedy_prob: float

@dataclass
class AgentConfig:
    steps: int
    time_limit: int
    k_fold_validation: int
    expose_prediction: bool
    data_preview: bool
    convert_system_to_user: bool
    obfuscate: bool
    retriever: StrategyConfig
    analyzer: StrategyConfig
    planner: StrategyConfig
    coder: StrategyConfig
    verifier: StrategyConfig
    improver: StrategyConfig
    search: SearchConfig

@dataclass
class ExecConfig:
    timeout: int
    agent_file_name: str
    format_tb_ipython: bool

@dataclass
class Config:
    data_dir: Path
    desc_file: Path | None
    goal: str | None
    eval: str | None
    log_dir: Path
    log_level: str
    workspace_dir: Path
    preprocess_data: bool
    copy_data: bool
    exp_name: str
    exec: ExecConfig
    generate_report: bool
    report: StageConfig
    agent: AgentConfig

# Create temporary directories
temp_dir = tempfile.mkdtemp()
data_dir = Path(temp_dir) / "data"
log_dir = Path(temp_dir) / "logs"
workspace_dir = Path(temp_dir) / "workspace"

# Create directories
data_dir.mkdir(exist_ok=True)
log_dir.mkdir(exist_ok=True)
workspace_dir.mkdir(exist_ok=True)

# Create a temporary task description file
desc_file = Path(temp_dir) / "task_desc.md"
with open(desc_file, 'w') as f:
    f.write("# Task Description\nThis is a test task.")

# Build config objects
search_config = SearchConfig(
    max_debug_depth=5,
    debug_prob=0.1,
    num_drafts=3,
    hpo_prob=0.2,
    trick_prob=0.15,
    max_hpo_prob=0.8,
    greedy_prob=0.5
)

exec_config = ExecConfig(
    timeout=300,
    agent_file_name="agent.py",
    format_tb_ipython=False
)

report_config = StageConfig(
    model="gpt-4",
    temp=0.7
)

strategy_config = StrategyConfig(
    strategy="default",
    model="gpt-4"
)

agent_config = AgentConfig(
    steps=10,
    time_limit=3600,
    k_fold_validation=5,
    expose_prediction=True,
    data_preview=True,
    convert_system_to_user=False,
    obfuscate=False,
    retriever=strategy_config,
    analyzer=strategy_config,
    planner=strategy_config,
    coder=strategy_config,
    verifier=strategy_config,
    improver=strategy_config,
    search=search_config
)

config = Config(
    data_dir=data_dir,
    desc_file=desc_file,
    goal=None,
    eval=None,
    log_dir=log_dir,
    log_level="INFO",
    workspace_dir=workspace_dir,
    preprocess_data=False,
    copy_data=False,
    exp_name="test_exp",
    exec=exec_config,
    generate_report=True,
    report=report_config,
    agent=agent_config
)

ARGS = {
    "cfg": config
}
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
opt_messages_to_list Ran
ran on a synthesized input
innovatingai/automind/automind/backend/utils.py
code served (permissive licence) · get_code("aa54a6716510580d")
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 torch

# The function takes two optional string parameters
system_message = "You are a helpful assistant."
user_message = "What is the capital of France?"

ARGS = {
    'system_message': system_message,
    'user_message': user_message
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — 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
zero_report Ran
ran; our contract draft was wrong, not the code
openai/mle-bench/experiments/aggregate_grading_reports.py
pointer only (licence: NOASSERTION) · get_code("0e40cd188f75fbbd")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, openai/mle-bench (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 7 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.

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

We introduce MLE-bench, a benchmark for measuring how well AI agents perform at machine learning engineering. To this end, we curate 75 ML engineeringrelated competitions from Kaggle, creating a diverse set of challenging tasks that test real-world ML engineering skills such as training models, preparing datasets, and running experiments. We establish human baselines for each competition using Kaggle's publicly available leaderboards. We use open-source agent scaffolds to evaluate several frontier language models on our benchmark, finding that the best-performing setup -OpenAI's o1-preview with AIDE scaffoldingachieves at least the level of a Kaggle bronze medal in 16.9% of competitions. In addition to our main results, we investigate various forms of resource-scaling for AI agents and the impact of contamination from pre-training. We open-source our benchmark code (github.com/openai/mle-bench/) to facilitate future research in understanding the ML engineering capabilities of AI agents.

For agents

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

get_harvested_code_for_paper("2410.07095")
get_code_for_paper("2410.07095")
have("2410.07095")

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

curl https://syntology.ai/api/ran/2410.07095.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.07095.svg)](https://syntology.ai/paper/2410.07095)

Connect an agent — have() is free.