SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2412.00535 · 2024

FullStack Bench: Evaluating LLMs as Full Stack Coders

arXiv · PDF · Open in the Atlas

Code that ran

13 of 14 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 0 have no contract verdict; 13 failures of our instrument; 1 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/2412.00535.json · badge: /api/ran/2412.00535.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
bytedance/sandboxfusion canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
bytedance/fullstackbench canonical (classifier label)
classifier basis: census
3 ran · 0 no recorded run
FunctionStatusWhere it lives
append_test Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/autoeval.py
code served (permissive licence) · get_code("0c8ec0958e7449cb")
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

torch.manual_seed(0)

# Analyze the function:
# - code: str - a string of Python code
# - test: str - a string containing test code, may have '#<INSERT>' token
# - repr_code: bool - whether to repr the code string

# The function:
# 1. Checks if '#<INSERT>' is in test
# 2. If yes: splits test by that token and inserts code (optionally repr'd) between parts
# 3. If no: removes 'if __name__ ==' block from code if present, then concatenates code + '\n' + test
# 4. Returns the full_code string

# Simple test case values
code_sample = """def hello():
    return "world"

result = hello()
print(result)"""

test_sample_with_insert = """assert result == "world"  #<INSERT>"""

test_sample_without_insert = """assert result == "world" """

ARGS = {
    "code": code_sample,
    "test": test_sample_with_insert,
    "repr_code": False
}
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
configurable_retry Ran
ran; our contract draft was wrong, not the code
bytedance/fullstackbench/src/utils.py
code served (permissive licence) · get_code("1f97a0c8318ed187")
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 function. 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 asyncio
from functools import wraps
from tenacity import retry, stop_after_attempt, wait_exponential_jitter

logging.basicConfig(level=logging.DEBUG)
logger = logging.getLogger(__name__)

def on_retry_error(s):
    e = s.outcome.exception()
    logger.error(f'give up retrying. error: {e}')
    raise e

def before_retry_sleep(s):
    msg = f'function call error for {s.attempt_number} time(s), will retry... error: {s.outcome.exception()}'
    if s.attempt_number > 2:
        logger.warning(msg)
    else:
        logger.debug(msg)

def configurable_retry(max_attempts):
    def decorator(func):
        @wraps(func)
        @retry(wait=wait_exponential_jitter(), stop=stop_after_attempt(max_attempts), before_sleep=before_retry_sleep, retry_error_callback=on_retry_error)
        async def async_wrapper(*args, **kwargs):
            return await func(*args, **kwargs)

        @wraps(func)
        @retry(wait=wait_exponential_jitter(), stop=stop_after_attempt(max_attempts), before_sleep=before_retry_sleep, retry_error_callback=on_retry_error)
        def sync_wrapper(*args, **kwargs):
            return func(*args, **kwargs)
        return async_wrapper if asyncio.iscoroutinefunction(func) else sync_wrapper
    return decorator


# Test with a simple synchronous function
@configurable_retry(max_attempts=3)
def sync_test_func(x):
    return x * 2


# Test with a simple asynchronous function
@configurable_retry(max_attempts=2)
async def async_test_func(x):
    await asyncio.sleep(0.01)
    return x * 3


ARGS = {
    "max_attempts": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tenacity — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
generate_fewshot_prompt Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/live_code_bench.py
code served (permissive licence) · get_code("88b93ad818bda02f")
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.

ARGS = {
    "question": "Given a list of integers, find the maximum element.\n\nInput\n\nA list of integers separated by spaces.\n\nOutput\n\nOutput the maximum element.\n\nSample Input\n\n1 5 3 9 2\n\nSample Output\n\n9",
    "starter_code": 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 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
generate_input_prompt Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/cruxeval.py
code served (permissive licence) · get_code("a31e27f6de0a1ad8")
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.

ARGS = {
    'language': 'python',
    'example_list': [
        {
            'code': 'def f(x):\n    return x + 1\n\nprint(f(candidate(5)))',
            'inputs': '5'
        },
        {
            'code': 'def f(x):\n    return x * 2\n\nprint(f(candidate(3)))',
            'inputs': '3'
        }
    ],
    'code_snippet': 'def f(x):\n    return x ** 2\n\nprint(f(candidate(????)))'
}
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
generate_output_prompt Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/cruxeval.py
code served (permissive licence) · get_code("310d913d34ff3a84")
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

# Set random seed for reproducibility
torch.manual_seed(0)

# Define the arguments for generate_output_prompt
language = "python"

example_list = [
    {
        'code': 'x = 5\nassert x == ????',
        'outputs': '5'
    },
    {
        'code': 'def add(a, b):\n    return a + b\nassert add(2, 3) == ????',
        'outputs': '5'
    }
]

code_snippet = 'y = 10\nassert y == ????'

ARGS = {
    'language': language,
    'example_list': example_list,
    'code_snippet': code_snippet
}
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
get_categories Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/humanevoeval.py
code served (permissive licence) · get_code("c1ac2d1e1f9e2e6d")
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 defaultdict. 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 re
from collections import defaultdict

class MockResult:
    def __init__(self, task_id):
        self.id = task_id

results = [
    MockResult("EvoEval_math_EvoEval"),
    MockResult("EvoEval_coding_EvoEval"),
    MockResult("EvoEval_math_EvoEval"),
    MockResult("EvoEval_logic_EvoEval"),
]

ARGS = {
    "results": results
}
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 defaultdict — a defaultdict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a defaultdict 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_table_name Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/database.py
code served (permissive licence) · get_code("42dedaf330d28e13")
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

ARGS = {
    'name': 'my_table',
    'db_type': 'sqlite'
}
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
max_concurrency Ran
ran; our contract draft was wrong, not the code
bytedance/fullstackbench/src/utils.py
code served (permissive licence) · get_code("5f94c3a946bcfdf6")
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 function. 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 asyncio
from functools import wraps

def max_concurrency(limit: int):
    """ Decorator to limit the maximum number of concurrent executions of an async function """
    semaphore = asyncio.Semaphore(limit)

    def decorator(func):

        @wraps(func)
        async def wrapper(*args, **kwargs):
            async with semaphore:
                return await func(*args, **kwargs)
        return wrapper
    return decorator


# Setup code to construct valid arguments
ARGS = {
    "limit": 3
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
postprocess_full_code Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/autoeval.py
code served (permissive licence) · get_code("cb97c663edcbd66d")
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 re

def postprocess_full_code(code, language):
    if language in ['go', 'go_test']:
        packages = set(re.findall('package\\s+(\\w+)', code))
        code = re.sub('package\\s+(\\w+)', '', code)
        single_imports = re.findall('import\\s+("\\w+")', code)
        multi_imports = sum([x.split('\n') for x in re.findall('import\\s+\\((.*?)\\)', code, flags=re.DOTALL)], [])
        multi_imports = [x.strip() for x in multi_imports]
        imports = set(single_imports + multi_imports) - {''}
        code = re.sub('import\\s+("\\w+")', '', code)
        code = re.sub('import\\s+\\((.*?)\\)', '', code, flags=re.DOTALL)
        code = '\n'.join([f'package {p}' for p in packages]) + '\n\n' + '\n'.join([f'import {imp}' for imp in imports]) + '\n\n' + code
    return code


# Test case 1: Go code with package and imports
go_code_sample = '''package main

import "fmt"
import (
    "os"
    "sys"
)

func main() {
    fmt.Println("Hello")
}
'''

# Test case 2: Non-go language
python_code_sample = '''def hello():
    print("world")
'''

ARGS = {
    "code": go_code_sample,
    "language": "go"
}
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
postprocess_full_code Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/humaneval.py
code served (permissive licence) · get_code("588673a8052153a0")
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 re

# The function takes code (string) and language (string) as parameters
# It processes Go code by extracting and reorganizing package/import declarations

ARGS = {
    "code": """package main

import "fmt"
import (
    "os"
    "io"
)

func main() {
    fmt.Println("Hello, World!")
}
""",
    "language": "go"
}
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
read_jsonl Ran
ran; our contract draft was wrong, not the code
bytedance/fullstackbench/src/utils.py
code served (permissive licence) · get_code("ff3dacdd2cf428ba")
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 json
import tempfile
import os

# Create a temporary JSONL file with sample data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.jsonl")

# Write sample JSONL data
with open(temp_file, 'w') as f:
    f.write(json.dumps({"id": 1, "text": "hello"}) + "\n")
    f.write(json.dumps({"id": 2, "text": "world"}) + "\n")
    f.write(json.dumps({"id": 3, "data": [1, 2, 3]}) + "\n")

ARGS = {
    "fn": temp_file
}
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
stop_token_trim Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/humaneval.py
code served (permissive licence) · get_code("52d455f13017fb44")
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 re
from typing import Any, Dict, List

# Construct test arguments for stop_token_trim
s = "Hello world\n\nThis is the end\n\nMore text"

row = {
    'stop_tokens': [
        '\n\n',           # in excepts, should be skipped
        '###',            # literal string stop token
        're:end.*',       # regex pattern stop token
    ]
}

excepts = ['\n\n']

ARGS = {
    's': s,
    'row': row,
    'excepts': excepts,
}
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
stop_token_trim Ran
ran; our contract draft was wrong, not the code
bytedance/sandboxfusion/sandbox/datasets/humaneval_deprecated.py
code served (permissive licence) · get_code("dcd915834ca5db10")
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

ARGS = {
    's': 'Hello world this is a test string',
    'row': {
        'labels': {
            'stop_tokens': ['world', 'test']
        }
    }
}
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
extract_question Not yet run
unverified (harvested, no recorded run)
bytedance/sandboxfusion/sandbox/datasets/live_code_bench.py
code served (permissive licence) · get_code("cb2a2f189c014acf")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 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 14 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

As the capabilities of code large language models (LLMs) continue to expand, their applications across diverse code intelligence domains are rapidly increasing. However, most existing datasets only evaluate limited application domains. To address this gap, we have developed a comprehensive code evaluation dataset FullStack Bench focusing on full-stack programming, which encompasses a wide range of application domains (e.g., basic programming, data analysis, software engineering, mathematics, and machine learning). Besides, to assess multilingual programming capabilities, in FullStack Bench, we design real-world instructions and corresponding unit test cases from 16 widely-used programming languages to reflect real-world usage scenarios rather than simple translations. Moreover, we also release an effective code sandbox execution tool (i.e., SandboxFusion) supporting various programming languages and packages to evaluate the performance of our FullStack Bench efficiently. Comprehensive experimental results on our FullStack Bench demonstrate the necessity and effectiveness of our FullStack Bench and SandboxFusion.

For agents

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

get_harvested_code_for_paper("2412.00535")
get_code_for_paper("2412.00535")
have("2412.00535")

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

curl https://syntology.ai/api/ran/2412.00535.json

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

[![Syntology run record](https://syntology.ai/api/ran/2412.00535.svg)](https://syntology.ai/paper/2412.00535)

Connect an agent — have() is free.