SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2408.08152 · 2024

DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search

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; 2 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 7 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/2408.08152.json · badge: /api/ran/2408.08152.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
deepseek-ai/deepseek-prover-v1.5 canonical (classifier label)
classifier basis not recorded on the link
6 ran · 0 no recorded run
augustepoiroux/LeanInteract canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
cot_prompt Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-prover-v1.5/prover/utils.py
code served (permissive licence) · get_code("77994fb797e63cb3")
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 = {
    'data': {
        'header': 'import Mathlib\nimport Aesop\n\nset_option maxHeartbeats 0\n\nopen BigOperators Real Nat Topology Rat\n\n',
        'informal_prefix': 'Theorem: The sum of two even numbers is even.\n',
        'formal_statement': 'theorem sum_of_evens (a b : ℕ) (ha : Even a) (hb : Even b) : Even (a + b) := by\n  sorry'
    }
}
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
extract_positions Ran
ran, honoured the contract we drafted
deepseek-ai/deepseek-prover-v1.5/prover/lean/ast_parser.py
code served (permissive licence) · get_code("4222bbd6373963f5")
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 2, dtype family int; 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

# Construct a nested node structure with position information
node = {
    'info': {
        'original': {
            'pos': 10,
            'endPos': 20
        }
    },
    'children': [
        {
            'info': {
                'original': {
                    'pos': 25,
                    'endPos': 35
                }
            }
        },
        {
            'data': 'some value',
            'nested': {
                'info': {
                    'original': {
                        'pos': 40,
                        'endPos': 50
                    }
                }
            }
        }
    ],
    'metadata': {
        'name': 'test_node'
    }
}

ARGS = {
    'node': node
}
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 2, dtype family int — 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
extract_vals Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-prover-v1.5/prover/lean/ast_parser.py
code served (permissive licence) · get_code("20ac111c42889614")
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)

# Analyzing the function extract_vals:
# - It takes a single parameter 'data'
# - It processes dict, list, and other types recursively
# - For dicts: checks if 'val' key exists and calls .strip() on it (expects string)
# - For dicts: iterates over .values()
# - For lists: iterates over items
# - Returns a list of strings

# Construct test data structure with nested dicts and lists containing strings
test_data = {
    'val': '  hello world  ',
    'nested_dict': {
        'val': '  nested string  ',
        'another_key': 'some value'
    },
    'nested_list': [
        {'val': '  item1  '},
        {'val': '  item2  '},
        'plain string'
    ],
    'deep': {
        'level': {
            'val': '  deep value  ',
            'inner_list': [
                {'val': '  inner1  '},
                [
                    {'val': '  inner2  '}
                ]
            ]
        }
    }
}

ARGS = {
    'data': test_data
}
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
get_project_lean_version Ran
ran; our contract draft was wrong, not the code
augustepoiroux/LeanInteract/src/lean_interact/utils.py
code served (permissive licence) · get_code("ff8bbf72d7f2fdcc")
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 tempfile
from pathlib import Path

# Create a temporary directory with a lean-toolchain file
temp_dir = tempfile.mkdtemp()
project_dir = Path(temp_dir)

# Create a lean-toolchain file with a valid Lean version
toolchain_file = project_dir / 'lean-toolchain'
with open(toolchain_file, 'w', encoding='utf-8') as f:
    f.write('leanprover/lean4:v4.0.0')

ARGS = {
    'project_dir': project_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 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_total_memory_usage Ran
ran, honoured the contract we drafted
augustepoiroux/LeanInteract/src/lean_interact/utils.py
code served (permissive licence) · get_code("f576abc866b698be")
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 int, ndim 0, dtype family int, shape () 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 psutil
import os

# The function expects a psutil.Process object
# We'll use the current process as a valid example
current_process = psutil.Process(os.getpid())

ARGS = {
    "proc": current_process
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
psutil — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, 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
message_intersects_code Ran
ran, violated the contract we drafted
augustepoiroux/LeanInteract/src/lean_interact/interface.py
code served (permissive licence) · get_code("553538799f029812")
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 bool, ndim 0, dtype family float, shape () 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
from typing import Literal

# Define the position and message classes as given
class Pos:
    def __init__(self, line: int, column: int):
        self.line = line
        self.column = column
    
    def __le__(self, other: 'Pos') -> bool:
        if self.line < other.line:
            return True
        if self.line == other.line:
            return self.column <= other.column
        return False
    
    def __lt__(self, other: 'Pos') -> bool:
        return self <= other and (not self == other)
    
    def __eq__(self, other):
        if not isinstance(other, Pos):
            return False
        return self.line == other.line and self.column == other.column

class Message:
    def __init__(self, start_pos: Pos, end_pos: Pos | None = None, 
                 severity: Literal['error', 'warning', 'info', 'trace'] = 'error',
                 data: str = ''):
        self.start_pos = start_pos
        self.end_pos = end_pos
        self.severity = severity
        self.data = data

class Sorry:
    def __init__(self, start_pos: Pos | None = None, end_pos: Pos | None = None,
                 goal: str = '', proof_state: int | None = None):
        self.start_pos = start_pos
        self.end_pos = end_pos
        self.goal = goal
        self.proof_state = proof_state

# Create test instances
pos1 = Pos(line=1, column=5)
pos2 = Pos(line=3, column=10)
pos3 = Pos(line=5, column=0)
pos4 = Pos(line=7, column=15)

# Test with Message
msg = Message(start_pos=pos1, end_pos=pos2, severity='error', data='test error')

# Test with positions
start_pos = Pos(line=2, column=0)
end_pos = Pos(line=6, column=0)

ARGS = {
    'msg': msg,
    'start_pos': start_pos,
    'end_pos': end_pos
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pydantic, torch — 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
non_cot_few_shot_prompt Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-prover-v1.5/prover/utils.py
code served (permissive licence) · get_code("23c78fbc0be837a2")
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)

LEAN4_DEFAULT_HEADER = 'import Mathlib\nimport Aesop\n\nset_option maxHeartbeats 0\n\nopen BigOperators Real Nat Topology Rat\n\n'

# Construct a minimal data dict that satisfies the function's requirements
data = {
    'formal_statement': 'theorem add_comm (a b : ℕ) : a + b = b + a := by',
    'formal_proof': '  ring'
}

# The function uses .get() with defaults for 'header' and 'informal_prefix'
# and directly accesses 'formal_statement' and 'formal_proof'
# So we provide all four keys to test the full behavior

ARGS = {
    'data': data
}
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
non_cot_prompt Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-prover-v1.5/prover/utils.py
code served (permissive licence) · get_code("e4ed28dc52c5179d")
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 = {
    'data': {
        'header': 'import Mathlib\nimport Aesop\n\nset_option maxHeartbeats 0\n\nopen BigOperators Real Nat Topology Rat\n\n',
        'informal_prefix': 'Prove that the sum of two even numbers is even:\n',
        'formal_statement': 'theorem sum_even (a b : ℕ) (ha : Even a) (hb : Even b) : Even (a + b) := by\n  sorry'
    }
}
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
parse_lean_version Ran
ran; our contract draft was wrong, not the code
augustepoiroux/LeanInteract/src/lean_interact/utils.py
code served (permissive licence) · get_code("eda86991d7c9e040")
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 = {
    'lean_version': 'leanprover/lean4:v4.0.0'
}
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
process_lean_file Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-prover-v1.5/prover/lean/ast_parser.py
code served (permissive licence) · get_code("789c40fc9b050fa1")
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 re

line_break_regex = re.compile('(?<=\\n)')

def process_lean_file(file_contents, byte_idx_1, byte_idx_2):

    def get_line(lines, line_number):
        if 1 <= line_number <= len(lines):
            return lines[line_number - 1]
        else:
            raise IndexError('Line number out of range')

    def convert_pos(lines, byte_idx):
        num_bytes = [len(line.encode('utf-8')) for line in lines]
        n = 0
        for i, num_bytes_in_line in enumerate(num_bytes, start=1):
            n += num_bytes_in_line
            if n > byte_idx:
                line_byte_idx = byte_idx - (n - num_bytes_in_line)
                if line_byte_idx == 0:
                    return (i, 1)
                line = get_line(lines, i)
                m = 0
                for j, c in enumerate(line, start=1):
                    m += len(c.encode('utf-8'))
                    if m >= line_byte_idx:
                        return (i, j + 1)
        return (len(lines), len(lines[-1]))

    def extract_string_between_positions(lines, byte_idx_1, byte_idx_2):
        line_1, column_1 = convert_pos(lines, byte_idx_1)
        line_2, column_2 = convert_pos(lines, byte_idx_2)
        extracted_string = []
        if line_1 == line_2:
            substring = lines[line_1 - 1][column_1 - 1:column_2 - 1]
            extracted_string.append(substring)
        else:
            extracted_string.append(lines[line_1 - 1][column_1 - 1:])
            for line in range(line_1 + 1, line_2):
                extracted_string.append(lines[line - 1])
            extracted_string.append(lines[line_2 - 1][:column_2 - 1])
        return ''.join(extracted_string)

    def convert_line_col_to_char_idx(lines, line, col):
        char_idx = 0
        for i in range(line - 1):
            char_idx += len(lines[i])
        char_idx += col - 1
        return char_idx
    lines = line_break_regex.split(file_contents)
    line_1, column_1 = convert_pos(lines, byte_idx_1)
    line_2, column_2 = convert_pos(lines, byte_idx_2)
    extracted_string = extract_string_between_positions(lines, byte_idx_1, byte_idx_2)
    char_idx_1 = convert_line_col_to_char_idx(lines, line_1, column_1)
    char_idx_2 = convert_line_col_to_char_idx(lines, line_2, column_2)
    return (extracted_string, line_1, column_1, line_2, column_2, char_idx_1, char_idx_2)


file_contents = "hello world\nfoo bar\nbaz qux"
byte_idx_1 = 0
byte_idx_2 = 11

ARGS = {
    "file_contents": file_contents,
    "byte_idx_1": byte_idx_1,
    "byte_idx_2": byte_idx_2,
}
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 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

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, 6 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 6 later) and 4 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
cot_prompt 2 other papers
2 later
2502.07640 (2025-02)
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving
2507.11275 (2025-07)
FMC: Formalization of Natural Language Mathematical Competition Problems
extract_positions 2 other papers
2 later
2502.07640 (2025-02)
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving
2507.11275 (2025-07)
FMC: Formalization of Natural Language Mathematical Competition Problems
extract_vals 2 other papers
2 later
2502.07640 (2025-02)
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving
2507.11275 (2025-07)
FMC: Formalization of Natural Language Mathematical Competition Problems
non_cot_few_shot_prompt 2 other papers
2 later
2502.07640 (2025-02)
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving
2507.11275 (2025-07)
FMC: Formalization of Natural Language Mathematical Competition Problems
non_cot_prompt 2 other papers
2 later
2502.07640 (2025-02)
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving
2507.11275 (2025-07)
FMC: Formalization of Natural Language Mathematical Competition Problems
process_lean_file 2 other papers
2 later
2502.07640 (2025-02)
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving
2507.11275 (2025-07)
FMC: Formalization of Natural Language Mathematical Competition Problems

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 DeepSeek-Prover-V1.5, an open-source language model designed for theorem proving in Lean 4, which enhances DeepSeek-Prover-V1 by optimizing both training and inference processes. Pre-trained on DeepSeekMath-Base with specialization in formal mathematical languages, the model undergoes supervised fine-tuning using an enhanced formal theorem proving dataset derived from DeepSeek-Prover-V1. Further refinement is achieved through reinforcement learning from proof assistant feedback (RLPAF). Beyond the single-pass whole-proof generation approach of DeepSeek-Prover-V1, we propose RMaxTS, a variant of Monte-Carlo tree search that employs an intrinsic-reward-driven exploration strategy to generate diverse proof paths. DeepSeek-Prover-V1.5 demonstrates significant improvements over DeepSeek-Prover-V1, achieving new state-of-the-art results on the test set of the high school level miniF2F benchmark ($63.5\%$) and the undergraduate level ProofNet benchmark ($25.3\%$).

For agents

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

get_harvested_code_for_paper("2408.08152")
get_code_for_paper("2408.08152")
have("2408.08152")

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

curl https://syntology.ai/api/ran/2408.08152.json

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

[![Syntology run record](https://syntology.ai/api/ran/2408.08152.svg)](https://syntology.ai/paper/2408.08152)

Connect an agent — have() is free.