SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.07714 · 2024

StableToolBench: Towards Stable Large-Scale Benchmarking on Tool Learning of Large Language Models

arXiv · PDF · Open in the Atlas

Code that ran

11 of 15 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, 2 violated it and 3 have no contract verdict; 6 failures of our instrument; 4 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/2403.07714.json · badge: /api/ran/2403.07714.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
openbmb/toolbench canonical (classifier label)
classifier basis: census
5 ran · 4 no recorded run
thunlp-mt/stabletoolbench canonical (classifier label)
classifier basis: plan_row
3 ran · 0 no recorded run
pwc-1/Paper-10 community (classifier label)
classifier basis not recorded on the link
3 ran · 0 no recorded run
FunctionStatusWhere it lives
c2p_dynamic_expand Ran
ran on a synthesized input
pwc-1/Paper-10/sew_d/modeling_sew_d.py
code served (permissive licence) · get_code("2901dfbf4ce068d8")
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 Tensor, ndim 4, dtype family float, shape (2, 3, 4, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 torch.nn as nn

torch.manual_seed(0)

# Analyze function usage:
# - c2p_pos: called with .broadcast_to() method, needs shape attribute
# - query_layer: accessed with .shape[0], .shape[1], .shape[2]
# - relative_pos: accessed with .shape[-1]
# All are tensors

# Create distinct dimension sizes to avoid ambiguity
batch_size = 2
seq_len = 3
hidden_dim = 4
rel_pos_dim = 5

# c2p_pos needs to be broadcastable to [batch_size, seq_len, hidden_dim, rel_pos_dim]
# Start with shape that can broadcast (e.g., shape [1, 1, 1, rel_pos_dim])
c2p_pos = torch.randn(1, 1, 1, rel_pos_dim)

# query_layer with shape [batch_size, seq_len, hidden_dim]
query_layer = torch.randn(batch_size, seq_len, hidden_dim)

# relative_pos with last dimension rel_pos_dim
relative_pos = torch.randn(batch_size, seq_len, rel_pos_dim)

ARGS = {
    "c2p_pos": c2p_pos,
    "query_layer": query_layer,
    "relative_pos": relative_pos,
}
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 Tensor, ndim 4, dtype family float, shape (2, 3, 4, 5) — shape (2, 3, 4, 5) as a later run recorded it
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
change_name Ran
ran; our contract draft was wrong, not the code
thunlp-mt/stabletoolbench/server/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from OceannTwT/Tool-Planner, is pointer only (licence: NONE) · get_code("7b6b723a33f5ea2a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from OceannTwT/Tool-Planner, and GitHub detected no licence in that repository when it was checked. 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.

chat_completion_request Ran
ran; our contract draft was wrong, not the code
openbmb/toolbench/toolbench/inference/LLM/chatgpt_function_model.py
code served (permissive licence) · get_code("b57a77da7c53ac7a")
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 APIRemovedInV1. 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 json
import openai
from unittest.mock import Mock, patch

# Create a minimal mock for openai.ChatCompletion.create
# to avoid actual API calls
def mock_chat_completion(**kwargs):
    """Mock response that simulates OpenAI API response"""
    mock_response = {
        'id': 'chatcmpl-123',
        'object': 'chat.completion',
        'created': 1234567890,
        'model': kwargs.get('model', 'gpt-3.5-turbo-16k-0613'),
        'choices': [
            {
                'index': 0,
                'message': {
                    'role': 'assistant',
                    'content': 'Test response'
                },
                'finish_reason': 'stop'
            }
        ],
        'usage': {
            'prompt_tokens': 10,
            'completion_tokens': 5,
            'total_tokens': 15
        }
    }
    return mock_response

# Patch openai.ChatCompletion.create before defining ARGS
with patch('openai.ChatCompletion.create', side_effect=mock_chat_completion):
    pass

# Define ARGS with all required and optional parameters
ARGS = {
    'key': 'sk-test-key-12345',
    'messages': [
        {
            'role': 'user',
            'content': 'Hello, how are you?'
        },
        {
            'role': 'assistant',
            'content': 'I am doing well, thank you for asking.'
        }
    ],
    'functions': None,
    'function_call': None,
    'key_pos': None,
    'model': 'gpt-3.5-turbo-16k-0613',
    'stop': None,
    'process_id': 0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
openai, tenacity — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type APIRemovedInV1 — a APIRemovedInV1 output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a APIRemovedInV1 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
format_speech_generation_kwargs Ran
ran on a synthesized input
pwc-1/Paper-10/seamless_m4t/modeling_seamless_m4t.py
code served (permissive licence) · get_code("a110d4f6e1706068")
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, 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
import torch.nn as nn

torch.manual_seed(0)

# The function takes a single parameter 'kwargs' which is a dictionary
# It processes the dictionary by splitting keys based on 'text_' and 'speech_' prefixes
# and distributes values to kwargs_text and kwargs_speech

# Create a realistic kwargs dictionary with various types of arguments
kwargs_input = {
    # Unprefixed arguments that go to both
    "max_length": 100,
    "num_beams": 4,
    "temperature": 0.9,
    
    # Text-specific arguments
    "text_max_length": 150,
    "text_num_beams": 5,
    "text_early_stopping": True,
    
    # Speech-specific arguments
    "speech_max_length": 200,
    "speech_num_beams": 3,
    
    # decoder_input_ids - should only go to text
    "decoder_input_ids": torch.tensor([[1, 2, 3], [4, 5, 6]]),
    
    # More mixed arguments
    "top_p": 0.95,
    "speech_top_p": 0.85,
}

ARGS = {
    "kwargs": kwargs_input
}
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 tuple, ndim 1, dtype family float — a tuple 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
format_speech_generation_kwargs Ran
ran on a synthesized input
pwc-1/Paper-10/seamless_m4t_v2/modeling_seamless_m4t_v2.py
code served (permissive licence) · get_code("8b5f9e0a6ad43062")
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, 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

torch.manual_seed(0)

# The function format_speech_generation_kwargs takes a kwargs dict
# and splits it into kwargs_text and kwargs_speech based on prefixes.
# It uses: kwargs.items() iteration and string operations (startswith, slicing)

ARGS = {
    "kwargs": {
        "max_length": 100,
        "num_beams": 4,
        "decoder_input_ids": torch.tensor([[1, 2], [3, 4]]),
        "text_temperature": 0.7,
        "speech_top_p": 0.9,
        "top_k": 50,
    }
}
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 tuple, ndim 1, dtype family float — a tuple 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_compressed_list Ran
ran; our contract draft was wrong, not the code
openbmb/toolbench/toolbench/model/compression.py
code served (permissive licence) · get_code("709a780a94ca2670")
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
import torch.nn as nn

torch.manual_seed(0)

# Create a simple module with nested Linear layers
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
        self.nested = nn.Sequential(
            nn.Linear(2, 5),
            nn.ReLU(),
            nn.Linear(5, 1)
        )
    
    def forward(self, x):
        return x

module = SimpleModel()

ARGS = {
    'module': module,
    'prefix': ''
}
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
is_valid_json Ran
ran, violated the contract we drafted
thunlp-mt/stabletoolbench/server/main_mirrorapi.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from lkutschka/notation-matters, is pointer only (licence: NOASSERTION) · get_code("be8785d230bc7581")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from lkutschka/notation-matters, and GitHub could not classify the licence of that repository (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.

process_system_message Ran
ran; our contract draft was wrong, not the code
openbmb/toolbench/toolbench/utils.py
code served (permissive licence) · get_code("a73f07e24a8b8a85")
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 torch.nn as nn

# Define a minimal system message that contains the required substring
system_message = "You are an AI assistant. Respond with a function call to actually excute your step."

# Define a minimal functions list/dict
functions = [
    {
        "name": "example_function",
        "description": "An example function",
        "parameters": {
            "type": "object",
            "properties": {
                "arg1": {"type": "string"}
            }
        }
    }
]

ARGS = {
    "system_message": system_message,
    "functions": functions
}
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
raise_warning_for_incompatible_cpu_offloading_configuration Ran
ran, violated the contract we drafted
openbmb/toolbench/toolbench/model/model_adapter.py
code served (permissive licence) · get_code("0bf414a2b44babe6")
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 bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, ndim and shape 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 sys
import warnings

# Set up arguments for the function
ARGS = {
    "device": "cuda",
    "load_8bit": True,
    "cpu_offloading": True,
}
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 bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
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
standardize Ran
ran; our contract draft was wrong, not the code
thunlp-mt/stabletoolbench/server/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from OceannTwT/Tool-Planner, is pointer only (licence: NONE) · get_code("e6d1b09cb3624d3f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from OceannTwT/Tool-Planner, and GitHub detected no licence in that repository when it was checked. 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.

standardize_category Ran
ran; our contract draft was wrong, not the code
openbmb/toolbench/toolbench/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from lkutschka/notation-matters, is pointer only (licence: NOASSERTION) · get_code("04eed77ef8a3068b")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from lkutschka/notation-matters, and GitHub could not classify the licence of that repository (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.

compress Not yet run
unverified (harvested, no recorded run)
openbmb/toolbench/toolbench/model/compression.py
code served (permissive licence) · get_code("386536152177a379")
How it ran

no invocation recorded.

get_conv_template Not yet run
unverified (harvested, no recorded run)
openbmb/toolbench/toolbench/tool_conversation.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from thierry123454/tool-selection-bias, is pointer only (licence: NONE) · get_code("eab40e1435b85709")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from thierry123454/tool-selection-bias, and GitHub detected no licence in that repository when it was checked. 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. No invocation recorded.

get_gpu_memory Not yet run
unverified (harvested, no recorded run)
openbmb/toolbench/toolbench/utils.py
code served (permissive licence) · get_code("2f3cb5d0dcc18502")
How it ran

no invocation recorded.

load_compress_model Not yet run
unverified (harvested, no recorded run)
openbmb/toolbench/toolbench/model/compression.py
code served (permissive licence) · get_code("5f9d4d17fb70af2b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 11 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 2 the same month, 7 later) and 3 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
get_conv_template 69 other papers
20 dated earlier, 3 the same month, 41 later, 4 same year, order not determinable, 1 undated
1911.03705 (2019-11)
CommonGen: A Constrained Text Generation Challenge for Generative Commonsense Reasoning
2112.05682 (2021-12)
Self-attention Does Not Need $O(n^2)$ Memory
2204.07705 (2022-04)
SUPER-NATURALINSTRUCTIONS: Generalization via Declarative Instructions on 1600+ NLP Tasks
and 66 more not named here
raise_warning_for_incompatible_cpu_offloading_configuration 15 other papers
7 dated earlier, 3 the same month, 5 later
2112.05682 (2021-12)
Self-attention Does Not Need $O(n^2)$ Memory
2310.13023 (2023-10)
GraphGPT: Graph Instruction Tuning for Large Language Models
2312.00401 (2023-12)
VIoTGPT: Learning to Schedule Vision Tools in LLMs towards Intelligent Video Internet of Things
and 12 more not named here
get_gpu_memory 7 other papers
7 dated earlier
2305.10973 (2023-05)
Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold
2307.12981 (2023-07)
3D-LLM: Injecting the 3D World into Large Language Models
2308.03279 (2023-08)
UniversalNER: TARGETED DISTILLATION FROM LARGE LANGUAGE MODELS FOR OPEN NAMED ENTITY RECOGNITION
and 4 more not named here
standardize 4 other papers
4 later
2406.03807 (2024-06)
Tool-Planner: Task Planning with Clusters across Multiple Tools
2410.08197 (2024-10)
From Exploration to Mastery: Enabling LLMs to Master Tools via Self-Driven Interactions
2412.10198 (2024-12)
From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection
and 1 more not named here
change_name 2 other papers
2 later
2406.03807 (2024-06)
Tool-Planner: Task Planning with Clusters across Multiple Tools
2605.29676 (2026-05)
Notation Matters: A Benchmark Study of Token-Optimized Formats in Agentic AI Systems
compress 2 other papers
2 dated earlier
2312.00401 (2023-12)
VIoTGPT: Learning to Schedule Vision Tools in LLMs towards Intelligent Video Internet of Things
2401.08326 (2024-01)
RoTBench: A Multi-Level Benchmark for Evaluating the Robustness of Large Language Models in Tool Learning
get_compressed_list 2 other papers
2 dated earlier
2312.00401 (2023-12)
VIoTGPT: Learning to Schedule Vision Tools in LLMs towards Intelligent Video Internet of Things
2401.08326 (2024-01)
RoTBench: A Multi-Level Benchmark for Evaluating the Robustness of Large Language Models in Tool Learning
load_compress_model 2 other papers
2 dated earlier
2312.00401 (2023-12)
VIoTGPT: Learning to Schedule Vision Tools in LLMs towards Intelligent Video Internet of Things
2401.08326 (2024-01)
RoTBench: A Multi-Level Benchmark for Evaluating the Robustness of Large Language Models in Tool Learning
standardize_category 2 other papers
2 later
2412.10198 (2024-12)
From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection
2605.29676 (2026-05)
Notation Matters: A Benchmark Study of Token-Optimized Formats in Agentic AI Systems
c2p_dynamic_expand 1 other paper
1 later
2406.18321 (2024-06)
MathOdyssey: Benchmarking Mathematical Problem-Solving Skills in Large Language Models Using Odyssey Math Data
is_valid_json 1 other paper
1 later
2605.29676 (2026-05)
Notation Matters: A Benchmark Study of Token-Optimized Formats in Agentic AI Systems

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

Large Language Models (LLMs) have witnessed remarkable advancements in recent years, prompting the exploration of tool learning, which integrates LLMs with external tools to address diverse real-world challenges. Assessing the capability of LLMs to utilise tools necessitates large-scale and stable benchmarks. However, previous works relied on either hand-crafted online tools with limited scale, or large-scale real online APIs suffering from instability of API status. To address this problem, we introduce StableToolBench, a benchmark evolving from ToolBench, proposing a virtual API server and stable evaluation system. The virtual API server contains a caching system and API simulators which are complementary to alleviate the change in API status. Meanwhile, the stable evaluation system designs solvable pass and win rates using GPT-4 as the automatic evaluator to eliminate the randomness during evaluation. Experimental results demonstrate the stability of StableToolBench, and further discuss the effectiveness of API simulators, the caching system, and the evaluator system.

For agents

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

get_harvested_code_for_paper("2403.07714")
get_code_for_paper("2403.07714")
have("2403.07714")

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

curl https://syntology.ai/api/ran/2403.07714.json

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

[![Syntology run record](https://syntology.ai/api/ran/2403.07714.svg)](https://syntology.ai/paper/2403.07714)

Connect an agent — have() is free.