LLM 评测榜单上榜指南

生成时间:2026-08-29 · 数据来源:benchmarks/ 各基准目录 README.md

全榜单速查总表

# 基准 提交方式 费用 奖金 联系方式
1SWE-bench VerifiedFork experiments 仓库开 PR / sb-cli submit免费无leaderboards@scale.com
2SWE-bench ProPublic 自测提交;Held-out/Commercial 官方代评Public 免费;代评洽谈无官网表单
3Terminal-Bench 2.1harbor run 后提交 Leaderboard PR免费无alexgshaw64@gmail.com
4Terminal-Bench HardAA 第三方统一评测(无独立入口)免费上榜无artificialanalysis.ai
5FrontierMath邮件申请私有集封闭评测洽谈人类赛 $10Kmath_evals@epoch.ai
6DeepSWEpier 运行后邮件提交公开版免费无serena@datacurve.ai
7FrontierSWE本地运行 / Prime CLI / 邮件免费无justus@proximal.ai
8Arena 三榜接 API 参与在线盲测投票见 rate card无arena.ai
9AA-OmniscienceAA 官方 API 自测Public 免费无artificialanalysis.ai
10HLE官方 harness 自测后邮件/PR免费$500K 已截止agibenchmark@safe.ai
11LiveBench本地自测后邮件/PR免费无livebenchai@gmail.com
12ARC-AGI-3Kaggle 代码竞赛 + 官方 API免费$850Karcprize.org

1. SWE-bench Verified 软件工程免费

官网https://www.swebench.com/
GitHubhttps://github.com/SWE-bench/SWE-bench
HuggingFaceprinceton-nlp/SWE-bench_Verified
数据规模500 题(人工验证子集;全量 SWE-bench 12000+ 题)
LicenseMIT · 无需申请
数据集下载HuggingFace · hf download princeton-nlp/SWE-bench_Verified --repo-type dataset
gtm 镜像1 文件 · 2.0 MB → 查看下载链接 ↓ · 打包下载 1.6 MB
论文SWE-bench: Can Language Models Resolve Real-world Github Issues? — Jimenez et al., ICLR 2024

上榜方式 / 提交方式

费用与奖金

联系方式

2. SWE-bench Pro 软件工程 · 防污染Public 免费Held-out 需申请

官网https://scaleapi.github.io/SWE-bench_Pro-os
GitHubhttps://github.com/scaleapi/SWE-bench_Pro-os
HuggingFaceScaleAI/SWE-bench_Pro
数据规模1865 题:Public 731 / Held-out 858 / Commercial 276
LicensePublic 子集 MIT;Held-out/Commercial 需申请授权
数据集下载HuggingFace(Public 731 题)· hf download ScaleAI/SWE-bench_Pro --repo-type dataset
gtm 镜像1 文件 · 7.5 MB → 查看下载链接 ↓ · 打包下载 5.7 MB
论文SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks? — Deng et al., 2025

上榜方式 / 提交方式

费用与奖金

联系方式

3. Terminal-Bench 2.1 终端智能体免费

官网https://www.tbench.ai/
GitHubhttps://github.com/harbor-framework/terminal-bench-2-1
HuggingFace无(数据在 GitHub 仓库内)
数据规模89 个任务(Harbor 格式)
LicenseApache-2.0 · 无需申请
数据集下载GitHub · git clone --depth 1 https://github.com/harbor-framework/terminal-bench-2-1.git
gtm 镜像89 任务目录 · 61 MB → 查看下载链接 ↓ · 打包下载 44 MB

上榜方式 / 提交方式

费用与奖金

联系方式

4. Terminal-Bench Hard 终端智能体 · 高难度第三方评测

官网https://artificialanalysis.ai/
GitHub无独立仓库(数据托管于 HF)
HuggingFaceZhongzhi1228/Terminal-Bench-Hard
数据规模100 题
LicenseCC-BY-4.0 · 无需申请
数据集下载HuggingFace · hf download Zhongzhi1228/Terminal-Bench-Hard --repo-type dataset
gtm 镜像100 任务 + fixtures → 查看下载链接 ↓ · 打包下载 169 KB

上榜方式 / 提交方式

费用与奖金

联系方式

5. FrontierMath 前沿数学私有集需申请

官网https://epoch.ai/frontiermath
GitHub无公开仓库(仅博客与评测页)
HuggingFace无公开数据集
数据规模全量约 350 题(公开样例 12 题)
License未明确开源(题目版权归 Epoch AI,私有集需申请)
数据下载Open Problems 数据(公开样例)· epoch.ai/open-problems
gtm 镜像4 文件 · 147 KB → 查看下载链接 ↓ · 打包下载 74 KB

上榜方式 / 提交方式

费用与奖金

联系方式

6. DeepSWE 深度软件工程公开版免费

官网https://deepswe.datacurve.ai/
GitHubhttps://github.com/datacurve-ai/deep-swe
HuggingFacedatacurve/deep-swe(gated)
数据规模113 个任务
LicenseApache-2.0(GitHub 公开版);HF gated 全量版需申请
数据集下载GitHub · git clone --depth 1 https://github.com/datacurve-ai/deep-swe.git · HuggingFace(gated)
gtm 镜像113 任务目录 · 34 MB → 查看下载链接 ↓ · 打包下载 2.0 MB

上榜方式 / 提交方式

费用与奖金

联系方式

7. FrontierSWE 前沿软件工程免费

官网https://frontierswe.com/
GitHubhttps://github.com/Proximal-Labs/frontier-swe
HuggingFace无
数据规模17 个任务
License见仓库(README/SCORING.md)
数据集下载GitHub · git clone --depth 1 https://github.com/Proximal-Labs/frontier-swe.git
gtm 镜像17 任务目录 · 935 MB → 查看下载链接 ↓ · 打包下载 286 MB

上榜方式 / 提交方式

费用与奖金

联系方式

8. Arena Leaderboards(事实性 / Expert / Occupational)在线盲测投票见 rate card

官网https://arena.ai/
GitHubhttps://github.com/lmarena/arena-rank
HuggingFacelmarena-ai/leaderboard-dataset(榜单快照)
数据规模无静态统一题集(线上盲测投票产生排名)
LicenseCC-BY-4.0(leaderboard-dataset)
数据集下载HuggingFace(榜单快照)· hf download lmarena-ai/leaderboard-dataset --repo-type dataset
gtm 镜像23 维度 · 46 文件 · 107 MB → 查看下载链接 ↓ · 打包下载 65 MB

上榜方式 / 提交方式

费用与奖金

联系方式

9. AA-Omniscience 跨领域知识可靠性Public 免费全量需申请

官网https://artificialanalysis.ai/evaluations/omniscience
GitHub无独立仓库
HuggingFaceArtificialAnalysis/AA-Omniscience-Public
数据规模Public 600 题;全量 6000 题私有
LicenseApache-2.0(Public 集);全量私有集需申请
数据集下载HuggingFace(Public 600 题)· hf download ArtificialAnalysis/AA-Omniscience-Public --repo-type dataset
gtm 镜像3 文件 · 4.0 MB → 查看下载链接 ↓ · 打包下载 3.3 MB
论文Evaluating Cross-Domain Knowledge Reliability in Large Language Models — Artificial Analysis Team, 2025

上榜方式 / 提交方式

费用与奖金

联系方式

10. HLE(Humanity's Last Exam)高难度多模态gated 需申请

官网https://agi.safe.ai/
GitHubhttps://github.com/centerforaisafety/hle
HuggingFacecais/hle(gated)
数据规模2500 题(文本+多模态混合)
LicenseMIT,但禁止用于模型训练
数据集下载HuggingFace(gated,需登录同意条款)· huggingface-cli login && hf download cais/hle --repo-type dataset
gtm 镜像gated 未落盘 → 详情 ↓
论文Humanity's Last Exam — Center for AI Safety & Scale AI, 2025

上榜方式 / 提交方式

费用与奖金

联系方式

11. LiveBench 防污染评测免费

官网https://livebench.ai/
GitHubhttps://github.com/LiveBench/LiveBench
HuggingFacelivebench(各 category 数据集)
数据规模6 大类别约 23 任务(coding / math / reasoning / language / data_analysis / instruction_following)
LicenseApache-2.0 · 无需申请
数据集下载GitHub · HuggingFace(各 category 数据集)
gtm 镜像6 类别 · 16 文件 · 246 MB → 查看下载链接 ↓ · 打包下载 172 MB
论文LiveBench: A Challenging, Contamination-Free Benchmark for LLMs — arXiv 2406.19314

上榜方式 / 提交方式

费用与奖金

联系方式

12. ARC-AGI-3 抽象推理 · AGI 挑战需 API Key

官网https://arcprize.org/arc-agi/3
GitHubhttps://github.com/arcprize/ARC-AGI-3-Agents
HuggingFace无
数据规模无静态题集(官方 API 提供题目与评测)
License见 GitHub 仓库(评测框架开源)
评测框架GitHub · git clone --depth 1 https://github.com/arcprize/ARC-AGI-3-Agents.git(无静态题集,通过官方 API 评测)

上榜方式 / 提交方式

费用与奖金

联系方式

gtm 镜像文件下载表

以下文件均已镜像至 gtm 服务器,可直接通过 HTTPS 下载。基础 URL:https://benchmark-comparison.zenagi.ailiantang.top/benchmarks/

文件路径大小
SWE-bench Verified 1 个文件 · 2.0 MB · 打包 1.6 MB
data/test-00000-of-00001.parquet2.0 MB
SWE-bench Pro 1 个文件 · 7.5 MB · 打包 5.7 MB
data/test-00000-of-00001.parquet7.5 MB
FrontierMath 4 个文件 · 147.1 KB · 打包 74 KB
open_problems_data.zip39.1 KB
open_problems_data/README.md620 B
open_problems_data/open_problems_metadata.csv6.7 KB
open_problems_data/open_problems_prompts.csv100.7 KB
AA-Omniscience 4 个文件 · 4.0 MB · 打包 3.3 MB
AA-Omniscience_dataset_public.csv151.0 KB
AA_Omniscience__Evaluating_Cross_Domain_Knowledge_Reliability_in_Large_Language_Models.pdf3.7 MB
Omniscience_Index_Question_Breakdown.png163.8 KB
README.md11.2 KB
Terminal-Bench Hard 49 个文件 · 94.9 KB · 打包 169 KB
metadata/tasks.parquet131 B
tasks/tbh_task_0512598a64b84b5e/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_06c8c1d0a94f66cb/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_08074c298b71613e/environment/_fixtures/app/fixtures/oracle14.9 KB
tasks/tbh_task_1037cf7b06cf39a5/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_15ef3de120c29cb9/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_16dbe4606059be59/environment/_fixtures/app/fixtures/oracle14.9 KB
tasks/tbh_task_19b99556078771ee/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_1b92e614b05da93c/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_22dc33aab6297db3/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_2d313903cabb4f6c/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_2d93f49a359191de/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_302dd2bfccb0feaf/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_30ee1bad124b54e4/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_33cf08c2cb634768/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_454c321755003c76/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_5631b23768838bc1/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_6355e8eeedf2490f/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_729d03972163d49b/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_75c31bc5a09f370c/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_7fee0e182f2ca1cd/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_881547b08c4e160e/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_8b3a71d1064b7be0/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_8b8b442da63e257c/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_9e4d1b21072cfe6b/environment/_fixtures/app/fixtures/oracle14.9 KB
tasks/tbh_task_a1b8dbbe01e426bc/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_a3636f22fa17cc6d/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_a73beb6749d55568/environment/_fixtures/app/fixtures/oracle14.9 KB
tasks/tbh_task_a7f0b82a4f205775/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_aa7daf3eb00fdcda/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_aaabe9842978fa41/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_aaf1a7faee31e564/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_aba8e19f775fc60e/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_abd213d6446d851a/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_b237a186590c87dd/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_b2c4085f2b2cd6a6/environment/_fixtures/app/fixtures/oracle14.9 KB
tasks/tbh_task_b528f72dbb214e70/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_c6c00a4dab8d28fb/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_d557a114712de69f/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_d574621170ea30f0/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_d575b3ecf9318243/environment/_fixtures/app/fixtures/oracle14.9 KB
tasks/tbh_task_d7a165b43c1bc0a1/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_d8154e00abf9ca0c/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_e7bd73bfecdd68d2/environment/_fixtures/app/fixtures/audio.wav130 B
tasks/tbh_task_f1d868d26a954722/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_f230e87093ffa566/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_f83d84c7dd0cc992/environment/_fixtures/app/fixtures/image.png129 B
tasks/tbh_task_fa699411e024a213/environment/_fixtures/app/fixtures/video.mp4129 B
tasks/tbh_task_fb2da3e652c57ac8/environment/_fixtures/app/fixtures/image.png129 B
Arena Leaderboards 44 个文件 · 106.4 MB · 打包 65 MB
agent/full-00000-of-00001.parquet40.8 KB
agent/latest-00000-of-00001.parquet8.0 KB
agent_bash_recovery_steps/full-00000-of-00001.parquet35.5 KB
agent_bash_recovery_steps/latest-00000-of-00001.parquet7.6 KB
agent_praise_complaint/full-00000-of-00001.parquet35.2 KB
agent_praise_complaint/latest-00000-of-00001.parquet7.6 KB
agent_steerability/full-00000-of-00001.parquet35.6 KB
agent_steerability/latest-00000-of-00001.parquet7.6 KB
agent_task_outcome_explicit/full-00000-of-00001.parquet35.5 KB
agent_task_outcome_explicit/latest-00000-of-00001.parquet7.6 KB
agent_tool_hallucination/full-00000-of-00001.parquet35.8 KB
agent_tool_hallucination/latest-00000-of-00001.parquet7.5 KB
document/full-00000-of-00001.parquet22.1 KB
document/latest-00000-of-00001.parquet7.0 KB
document_style_control/full-00000-of-00001.parquet22.1 KB
document_style_control/latest-00000-of-00001.parquet7.0 KB
image_edit/full-00000-of-00001.parquet104.4 KB
image_edit/latest-00000-of-00001.parquet9.9 KB
image_to_video/full-00000-of-00001.parquet59.4 KB
image_to_video/latest-00000-of-00001.parquet7.6 KB
search/full-00000-of-00001.parquet36.4 KB
search/latest-00000-of-00001.parquet6.8 KB
search_factuality/full-00000-of-00001.parquet9.2 KB
search_factuality/latest-00000-of-00001.parquet6.5 KB
search_style_control/full-00000-of-00001.parquet36.4 KB
search_style_control/latest-00000-of-00001.parquet6.8 KB
text/full-00000-of-00001.parquet51.8 MB
text/latest-00000-of-00001.parquet566.2 KB
text_factuality/full-00000-of-00001.parquet2.0 MB
text_factuality/latest-00000-of-00001.parquet172.6 KB
text_style_control/full-00000-of-00001.parquet45.5 MB
text_style_control/latest-00000-of-00001.parquet566.6 KB
text_to_image/full-00000-of-00001.parquet720.4 KB
text_to_image/latest-00000-of-00001.parquet34.3 KB
text_to_video/full-00000-of-00001.parquet58.0 KB
text_to_video/latest-00000-of-00001.parquet7.6 KB
video_edit/full-00000-of-00001.parquet7.4 KB
video_edit/latest-00000-of-00001.parquet5.5 KB
vision/full-00000-of-00001.parquet1.8 MB
vision/latest-00000-of-00001.parquet50.8 KB
vision_style_control/full-00000-of-00001.parquet1.7 MB
vision_style_control/latest-00000-of-00001.parquet50.8 KB
webdev/full-00000-of-00001.parquet815.1 KB
webdev/latest-00000-of-00001.parquet24.4 KB
LiveBench 16 个文件 · 241.4 MB · 打包 172 MB
assets/livebench-2024-07-20.png851.4 KB
assets/livebench-2024-07-23.png819.2 KB
assets/livebench-2024-07-24.png845.4 KB
assets/livebench-2024-07-28.png838.0 KB
assets/livebench-2024-08-02.png851.5 KB
assets/livebench-2024-08-06.png830.5 KB
assets/livebench-2024-08-30.png692.2 KB
assets/livebench-2024-09-30.png932.6 KB
assets/livebench-2024-12-01.png306.5 KB
coding/data/test-00000-of-00001.parquet233.4 MB
data_analysis/data/test-00000-of-00001.parquet141.4 KB
instruction_following/data/test-00000-of-00001.parquet524.4 KB
language/data/test-00000-of-00001.parquet281.1 KB
language/data/train-00000-of-00001.parquet131 B
math/data/test-00000-of-00001.parquet181.3 KB
reasoning/data/test-00000-of-00001.parquet86.2 KB
DeepSWE 113 个任务 · 打包 2.0 MB
abs-module-cache-flags/目录
abs-stepped-slices/目录
actionlint-action-pinning-lint/目录
adaptix-name-mapping-aliases/目录
aiomonitor-task-snapshots-diff/目录
anko-default-function-arguments/目录
anko-typed-variable-bindings/目录
arcane-drift-detection-baselines/目录
arktype-json-schema-refs-dependencies/目录
awilix-async-container-initialization/目录
bandit-incremental-cache-control/目录
bandit-interprocedural-taint-checks/目录
bandit-structured-nosec-directives/目录
boa-hierarchical-evaluation-cancellation/目录
cattrs-partial-structuring-recovery/目录
clack-async-autocomplete-options/目录
claude-code-by-agents-recursive-delegation/目录
cliffy-config-file-parsing/目录
csstree-shorthand-expansion-compression/目录
dasel-html-document-format/目录
dateutil-rfc5545-timezone-interop/目录
drizzle-orm-window-function-builders/目录
dynamodb-toolbox-conditional-attribute-requirements/目录
dynamodb-toolbox-lazy-recursive-schemas/目录
effect-sse-httpapi-streaming/目录
eicrud-keyset-pagination-cursor/目录
etree-xml-diff-patch/目录
expr-try-catch-errors/目录
fastapi-deprecation-response-headers/目录
fastapi-implicit-head-options/目录
fd-deterministic-multi-key-sorting/目录
geo-shapeindex-serialization/目录
go-critic-doc-link-checker/目录
go-genai-streamed-function-args/目录
go-git-worktree-merge-conflicts/目录
goreleaser-retry-publish-auditing/目录
gql-incremental-graphql-delivery/目录
happy-dom-abort-pending-body-reads/目录
happy-dom-deterministic-intersectionobserver/目录
helm-array-merge-strategies/目录
helm-unified-manifest-stream/目录
httpx-deterministic-cookie-store/目录
httpx-multipart-response-parsing/目录
httpx-streaming-json-iteration/目录
igel-persist-feature-schema/目录
ink-grid-box-layout/目录
ipython-session-bundle-replay/目录
katex-multicolumn-array-spans/目录
kcp-go-multiplexed-kcp-streams/目录
kea-atomic-signal-selectors/目录
kgateway-consistent-hash-policy/目录
kombu-single-active-consumer-priority/目录
kombu-virtual-queue-dead-lettering/目录
koota-composite-trait-aspects/目录
koota-deferred-mutation-buffer/目录
koota-entity-snapshot-rollback/目录
koota-pair-relation-tracking/目录
koota-query-predicates/目录
kysely-window-grouping-helpers/目录
langchain-request-coalescing/目录
mashumaro-flattened-dataclass-fields/目录
meriyah-explicit-resource-declarations/目录
mnamer-daemon-watch-lifecycle/目录
mobly-grouped-test-barriers/目录
narwhals-rolling-window-suite/目录
numba-stencil-boundary-modes/目录
obsidian-linter-auto-table-of-contents/目录
obsidian-linter-link-format-conversion/目录
obsidian-linter-scoped-ignore-markers/目录
ofetch-per-origin-circuit-breaker/目录
onedump-dump-encryption-pipeline/目录
opa-rego-rule-profiling/目录
opa-template-string-reconstruction/目录
optique-conditional-option-dependencies/目录
oxvg-structural-selector-preservation/目录
participle-grammar-conflict-analysis/目录
pebble-durability-wait-apis/目录
pest-character-class-coalescing/目录
prometheus-transactional-reload-status/目录
prometheus-typed-label-sorting/目录
psd-tools-blend-range-api/目录
pwntools-tube-multiplexing/目录
python-statemachine-state-data-scoping/目录
query-persist-restored-query-state/目录
quill-shared-toolbar-focus/目录
returns-validated-error-accumulation/目录
scc-bounded-memory-spilling/目录
scriggo-method-declarations/目录
skrub-duration-encoding/目录
sql-formatter-bigquery-pipe-formatting/目录
sqlfmt-create-table-ddl-formatting/目录
sqlite-utils-safe-import-checkpoints/目录
superjson-error-stack-serialization/目录
task-task-graph-export/目录
tengo-callable-instance-isolation/目录
tengo-destructuring-bindings/目录
termenv-preserve-ansi-resets/目录
testem-bail-on-test-failure/目录
testem-per-launcher-reports/目录
textual-kitty-key-phases/目录
textual-richlog-follow-state/目录
tomlkit-toml-table-converters/目录
true-myth-iterable-collection-combinators/目录
ts-pattern-match-each/目录
updo-policy-alerting/目录
valibot-recursive-schema-composition/目录
vitest-duration-sharding/目录
vulture-persistent-analysis-cache/目录
wasmi-trap-coredumps/目录
wazero-multi-module-snapshots/目录
yaegi-go-embed-directives/目录
yjs-map-conflict-detection/目录
ytt-jsonpath-query-api/目录
FrontierSWE 17 个任务 · 打包 286 MB
cranelift-codegen-opt/目录
dart-style-haskell/目录
dependent-type-checker/目录
ffmpeg-swscale-rewrite/目录
frogsgame-rl/目录
git-to-zig/目录
granite-mamba2-inference-optimization/目录
inference-system-optimization/目录
libexpat-to-x86asm/目录
lua-native-compiler/目录
modular-stack-wan21/目录
notebook-compression/目录
optimizer-design/目录
pcqm4mv2-autoresearch/目录
postgres-sqlite-wire-adapter/目录
pyright-type-checking-optimization/目录
revideo-perf-opt/目录
Terminal-Bench 2.1 89 个任务 · 打包 44 MB
adaptive-rejection-sampler/目录
bn-fit-modify/目录
break-filter-js-from-html/目录
build-cython-ext/目录
build-pmars/目录
build-pov-ray/目录
caffe-cifar-10/目录
cancel-async-tasks/目录
chess-best-move/目录
circuit-fibsqrt/目录
cobol-modernization/目录
code-from-image/目录
compile-compcert/目录
configure-git-webserver/目录
constraints-scheduling/目录
count-dataset-tokens/目录
crack-7z-hash/目录
custom-memory-heap-crash/目录
db-wal-recovery/目录
distribution-search/目录
dna-assembly/目录
dna-insert/目录
extract-elf/目录
extract-moves-from-video/目录
feal-differential-cryptanalysis/目录
feal-linear-cryptanalysis/目录
filter-js-from-html/目录
financial-document-processor/目录
fix-code-vulnerability/目录
fix-git/目录
fix-ocaml-gc/目录
gcode-to-text/目录
git-leak-recovery/目录
git-multibranch/目录
gpt2-codegolf/目录
headless-terminal/目录
hf-model-inference/目录
install-windows-3.11/目录
kv-store-grpc/目录
large-scale-text-editing/目录
largest-eigenval/目录
llm-inference-batching-scheduler/目录
log-summary-date-ranges/目录
mailman/目录
make-doom-for-mips/目录
make-mips-interpreter/目录
mcmc-sampling-stan/目录
merge-diff-arc-agi-task/目录
model-extraction-relu-logits/目录
modernize-scientific-stack/目录
mteb-leaderboard/目录
mteb-retrieve/目录
multi-source-data-merger/目录
nginx-request-logging/目录
openssl-selfsigned-cert/目录
overfull-hbox/目录
password-recovery/目录
path-tracing/目录
path-tracing-reverse/目录
polyglot-c-py/目录
polyglot-rust-c/目录
portfolio-optimization/目录
protein-assembly/目录
prove-plus-comm/目录
pypi-server/目录
pytorch-model-cli/目录
pytorch-model-recovery/目录
qemu-alpine-ssh/目录
qemu-startup/目录
query-optimize/目录
raman-fitting/目录
regex-chess/目录
regex-log/目录
reshard-c4-data/目录
rstan-to-pystan/目录
sam-cell-seg/目录
sanitize-git-repo/目录
schemelike-metacircular-eval/目录
sparql-university/目录
sqlite-db-truncate/目录
sqlite-with-gcov/目录
torch-pipeline-parallelism/目录
torch-tensor-parallelism/目录
train-fasttext/目录
tune-mjcf/目录
video-processing/目录
vulnerable-secret/目录
winning-avg-corewars/目录
write-compressor/目录
HLE gated 未落盘
数据为 gated 需 HF 登录授权,未落盘至服务器。
HF 数据集:huggingface.co/datasets/cais/hle(登录后即可下载)
仓库 README:github.com/centerforaisafety/hle
ARC-AGI-3 无静态数据集
无静态题集,通过官方 API 评测。评测框架代码见 GitHub。
回到总表