Explore

Datasets

Public task collections. Sign in to inspect tasks and configure an evaluation.

Loading records…

Data source

Source: Gateway public Registry

ale/rsi-post-training
Community catalog
Technical details
ALE RSI post-training benchmark tasks for evaluating agent capabilities in research-level post-training workflows
Dataset reference
harbor:ale/rsi-post-training@sha256:8f9485dd374141adf8bf8f6475eb6d107a59351ce585674586fb3175222bce4c
openthoughts/tasktrove-exp-rpt-unitsyn-python-v3
Community catalog
Technical details
exp rpt unitsyn-python-v3 tasks (500 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: DCAgent/exp_rpt_unitsyn-python-v3.
Dataset reference
harbor:openthoughts/tasktrove-exp-rpt-unitsyn-python-v3@sha256:cf0d2e6a48783b2a8bd9733dff7991685f23b099489ade649f6dd848f385c3ae
bigcode/humanevalfix
Community catalog
Technical details
Dataset reference
harbor:bigcode/humanevalfix@sha256:f3188245370b2ea142c6442591299c955202cfdf822371b4c6e9e5baf7dafe32
replicationbench/replicationbench
Community catalog
Technical details
Dataset reference
harbor:replicationbench/replicationbench@sha256:323a161d95ba985abfe8895fd6ca00bfb4233dad8a08adb6658e696bb2174f53
strongreject/strongreject
Community catalog
Technical details
Dataset reference
harbor:strongreject/strongreject@sha256:c3d584ac2b1b50436fe5c6e8f99ebef907cf6808747f36371f4975d1b9bc6b2f
silrlab/long-horizon-terminal-bench-2.0
Community catalog
Technical details
Long-Horizon Terminal-Bench 2.0: 77-task benchmark measuring how well LLM agents sustain useful work in a containerized terminal over hundreds of steps. The 46 original tasks plus 31 new ones, 30 of them reconstruction tasks. Hidden rebuild-from-artifact verifiers; self-reported progress does not count.
Dataset reference
harbor:silrlab/long-horizon-terminal-bench-2.0@sha256:7ab245d0ff3c7d0fdf40f8e8387bc3ed596cb23641bfba3746dd2fc30d84d3e6
elfsong/cpu2gpu
Community catalog
Technical details
CPU-to-GPU porting benchmark: 258 novel multicore CPU programs to port to CUDA, graded on hidden inputs for correctness and decades of speedup over a direct reference port
Dataset reference
harbor:elfsong/cpu2gpu@sha256:320024e75e721ef021e217d0a79cf7205404533e0fbc5a5efc2cf7d3e6510d56
arcus-labs/tau3-banking-dev
Community catalog
Technical details
T3 banking dev split (30 tasks); base image pulled from mirror.gcr.io instead of Docker Hub
Dataset reference
harbor:arcus-labs/tau3-banking-dev@sha256:1d7639f747a46d5d01c5330a3a31067fa14fd049bed963b931e404c27babba07
viratzzs/petct-pilot
Community catalog
Technical details
Private two-task PET/CT benchmark pilot; separately scored lesion mapping and result repair tasks.
Dataset reference
harbor:viratzzs/petct-pilot@sha256:57f739aeb396cf68e734010071986a476240903d4d47681d409d234ab209e800
blobfishai/factorybench-100
Community catalog
Technical details
100 executable manufacturing ERP workflow tasks with deterministic FactoryScore grading
Dataset reference
harbor:blobfishai/factorybench-100@sha256:9634a8bcb5962f922dbb3941e60f667ca174252fa7506e73e3d034946d3abd62
openthoughts/tasktrove-exp-rpt-e2egit-v2
Community catalog
Technical details
exp rpt e2egit-v2 tasks (500 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: DCAgent/exp_rpt_e2egit-v2.
Dataset reference
harbor:openthoughts/tasktrove-exp-rpt-e2egit-v2@sha256:b2476344118aeb51aaa1f8dd46f8e7ed9d361c34e53dc3894467050af161b66e
aime/aime
Community catalog
Technical details
Dataset reference
harbor:aime/aime@sha256:534167df6e50234cdc0aa9c40a7dc6a60f27aeffc18ac981577f5e3272da3589
elfsong/mercury-eval
Community catalog
Technical details
Mercury eval: 256 Python code-efficiency tasks; CC-BY-NC-4.0.
Dataset reference
harbor:elfsong/mercury-eval@sha256:4b312cd01387beb37547ebb217c57dc82b5fda1e098deee915a59aefbbf93efb
aarr/aarri-bench
Community catalog
Technical details
aarri(act as a real research intern)-bench, for evaluating LLM agents in academic research tasks
Dataset reference
harbor:aarr/aarri-bench@sha256:4cdb4aea9a63c16202e4b94b6f800ff2b3e0ba9969d70ca63c4289e185f20438
openthoughts/tasktrove-swegym-tasks-patched-validated-v2
Community catalog
Technical details
swegym-tasks-patched-validated-v2 tasks (989 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: laion/swegym-tasks-patched-validated-v2.
Dataset reference
harbor:openthoughts/tasktrove-swegym-tasks-patched-validated-v2@sha256:d216fe25a18b498ba86565d6d116135fcac837847d759b79af0d0ccf4ed2d1d5
abhishek203/task-aw
Community catalog
Technical details
IT support mock-provider tasks (a-1, a-2, a-3, a-4, a-8)
Dataset reference
harbor:abhishek203/task-aw@sha256:859248e86fb397605b06cbc53d89642eb1c160eec4aecd2d0a1778ba42b7f53f
bauerjustin/terminal-bench-3-test
Community catalog
Technical details
Terminal Bench 3 tasks
Dataset reference
harbor:bauerjustin/terminal-bench-3-test@sha256:6e510c131dbf307c3ddb4e0ede5e98a2bf383a15766c8e2139918f9e8d4488ce
vmax-modal/modal-port-v1-train-patched
Community catalog
Technical details
Patched verifier rebuild: reworked Modal capability-contract checks + unified reward (behavior x contract_gate). Tasks identical except tests/verify.py.
Dataset reference
harbor:vmax-modal/modal-port-v1-train-patched@sha256:d5ec77c57849075137a00af468430a98529af2f3cc149ae96fdf1d5121b35ee7
terminal-bench-pro/terminal-bench-pro
Community catalog
Technical details
Dataset reference
harbor:terminal-bench-pro/terminal-bench-pro@sha256:1e0df73902c63a708b75e5a4b633eaa21524583904fdadaf937203280f037461
swe-bench/swe-bench-verified
Community catalog
Technical details
Dataset reference
harbor:swe-bench/swe-bench-verified@sha256:b934b0cc3dc800fe945eaf9f1623329db97ee3133c706d20644524c7759fb341
reasoning-gym/reasoning-gym-easy
Community catalog
Technical details
Dataset reference
harbor:reasoning-gym/reasoning-gym-easy@sha256:1e82e6a129de757f0cdca355db373131c5b5dc9d9bf03d1ceabea67fde5e57a9
swe-rebench/swe-rebench-leaderboard
Community catalog
Technical details
SWE-rebench leaderboard: 860 curated Python SWE tasks from Nebius AI R&D. All instances have pre-built Docker images. Continuously updated monthly splits.
Dataset reference
harbor:swe-rebench/swe-rebench-leaderboard@sha256:ebe7444e313a0d8db94fa541139826eaebe2b0abcd4900c6f73e750494910dca
islo-labs/reward-hack-bench-control
Community catalog
Technical details
Baseline/control companion to reward-hack-bench: the same 8 SWE-bench + CyBench tasks with NO cheat prompt, still scored by the fairness judge. Also packages these CyBench/GlacierCTF tasks as Harbor tasks.
Dataset reference
harbor:islo-labs/reward-hack-bench-control@sha256:99138aaa3bec4e5fccc5a3c63b547604320d70b754aafbb5389660c202f37332
openthoughts/tasktrove-exp-rpt-methods2test-large-v2
Community catalog
Technical details
exp rpt methods2test-large-v2 tasks (4472 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: laion/exp_rpt_methods2test-large-v2.
Dataset reference
harbor:openthoughts/tasktrove-exp-rpt-methods2test-large-v2@sha256:a1197087a6a424387f2fc6b1deb690abd65b2e336a1f67c9f1d7d86b43e20988
punitarani/harvey-labs
Community catalog
Technical details
Harvey Labs: 1,760 legal-drafting tasks (LAB) across 26 practice areas, graded with Harbor rewardkit (LLM-judge, OpenRouter default, 111,826 binary criteria).
Dataset reference
harbor:punitarani/harvey-labs@sha256:a5ffde4541a825f3917c35e81b48b5ff79145502386881f03c1af6c6d3aeab9f
thetalab/vector-edit-gym
Community catalog
Technical details
106 hand-authored SVG editing tasks across four difficulty tiers (easy / medium / hard / very_hard). Each task is a corrupted SVG plus a natural-language fix instruction. Strict structural and preservation metrics catch stylistic drift.
Dataset reference
harbor:thetalab/vector-edit-gym@sha256:b2ce5a813991c55c7b610c77a52a5c54f656aeb2b7239558a0a846a9be62c625
openthoughts/tasktrove-code-contests-noblock
Community catalog
Technical details
code-contests-noblock tasks (8728 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: DCAgent/code-contests-noblock.
Dataset reference
harbor:openthoughts/tasktrove-code-contests-noblock@sha256:b0e82f923debe69cc13f4197184b47f9644ac8a056b5a29c1596b45ec7b2425b
gabeorlanski/slopcodebench
Community catalog
Technical details
SlopCodeBench multi-checkpoint coding benchmark tasks converted for Harbor.
Dataset reference
harbor:gabeorlanski/slopcodebench@sha256:aec29354c19d3e762640ab6d7d3c63ba8fcf4895e98a5756aecb9157b5bb4ae0
itmo-autods/fdata-exit-case
Community catalog
Technical details
F-DATA Fugaku exit-state case (SciData 2025)
Dataset reference
harbor:itmo-autods/fdata-exit-case@sha256:6d68145ff02f46c1d43020378f930b108b1d44d00d57296447bfa2151d904744
bigcode/bigcodebench-hard-complete
Community catalog
Technical details
Dataset reference
harbor:bigcode/bigcodebench-hard-complete@sha256:4c881f46251c98f6af182ec8eedbacc2c144db0761fcbbd789a60d7e69c30f69
blobfishai/litigation
Community catalog
Technical details
Executable litigation environments for agent training and evaluation. 100 long-horizon tasks, a median of 108 verified tool-calling steps each, worked across eight systems of record (Clio Manage, iManage Work, CM/ECF, Google Calendar, Docusign, Relativity, CourtListener and nine litigation sub-systems). Graded by deterministic state-diff and trace assertions — no LLM judge on the reward path.
Dataset reference
harbor:blobfishai/litigation@sha256:53d0cd3433b743c42671d10ef1b4de874c447b4994020f702cb5a31e7bf1ba0c
openthoughts/tasktrove-swe-rebench-v2-patched-oracle
Community catalog
Technical details
swe rebench v2 patched oracle tasks (18341 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: DCAgent/swe_rebench_v2_patched_oracle.
Dataset reference
harbor:openthoughts/tasktrove-swe-rebench-v2-patched-oracle@sha256:d2b680ed9144d575e016ced0c6d82247aa5448cb453ef7cb92f3d2a93a1ddc0d
openthoughts/tasktrove-exp-rpt-pymethods2test-v3
Community catalog
Technical details
exp rpt pymethods2test-v3 tasks (500 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: DCAgent/exp_rpt_pymethods2test-v3.
Dataset reference
harbor:openthoughts/tasktrove-exp-rpt-pymethods2test-v3@sha256:29bd0ecc2c2a668c4ac1a46d016845134691056360ed3c72016c8a5b5f955b91
huu256/gauntlet
Community catalog
Technical details
Gauntlet: five hard fundraising-desk games (6x6 to 14x14), a stochastic job-shop scheduling problem played through a chat with a hidden game server. Pass/fail against a fixed bar.
Dataset reference
harbor:huu256/gauntlet@sha256:db0c4e0c58bfb86d5aaca936319bc67c62768955501f9b8b3c6b08e359f2057c
openthoughts/tasktrove-exp-rpt-stack-go-v4
Community catalog
Technical details
exp rpt stack-go-v4 tasks (2313 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: laion/exp_rpt_stack-go-v4.
Dataset reference
harbor:openthoughts/tasktrove-exp-rpt-stack-go-v4@sha256:0d47f8629c4815e599c3912e9a05f7e755b7af249b4b18c7f49d7f4258383d71
kumo/kumo-1
Community catalog
Technical details
Dataset reference
harbor:kumo/kumo-1@sha256:a70e4f6c1a7cca8977db77461c8d341fb8dc481c857a48e28bc2b4ddcfa9e0ef
crustbench/crustbench
Community catalog
Technical details
Dataset reference
harbor:crustbench/crustbench@sha256:a474b77ebbe183daf14e367e1b9d6e7aa148634a982bfb0e937944743f618243
billxbf/terminal-agent-mix
Community catalog
Technical details
9k mix of public terminal / coding agent data
Dataset reference
harbor:billxbf/terminal-agent-mix@sha256:bbcc57ddff630d6ef1fa2553518306ec4937bc856f61377ef3d48550c6f45933
scale-ai/swe-bench-pro
Community catalog
Technical details
Dataset reference
harbor:scale-ai/swe-bench-pro@sha256:88411d32ff27e53a4c1a7e29f0c2aeba180c8e5d60f221cab5ed56325f33549d
scienceagentbench/scienceagentbench
Community catalog
Technical details
ScienceAgentBench full benchmark adapted to Harbor with Modal GPU support for GPU-backed tasks.
Dataset reference
harbor:scienceagentbench/scienceagentbench@sha256:e1b96865e47796cdfa47afae7b9bab3d4cf0cbea7d699e085ce660b716a57041
billxbf/mimo-v26-music
Community catalog
Technical details
MiMo-V2.6-RL Music: 1000 Harbor tasks (converted from XiaomiMiMo/MiMo-V2.6-RL-oss)
Dataset reference
harbor:billxbf/mimo-v26-music@sha256:a03b0c2b4b9eac371da01f61892dd0adf874ff795297ce06f628718e6aed7c01
openthoughts/tasktrove-exp-rle-heavy-padding-v2
Community catalog
Technical details
exp rle heavy padding-v2 tasks (784 tasks). Part of OpenThoughts TaskTrove. See https://huggingface.co/datasets/open-thoughts/TaskTrove. Source: laion/exp_rle_heavy_padding-v2.
Dataset reference
harbor:openthoughts/tasktrove-exp-rle-heavy-padding-v2@sha256:85786c0a9bd90fc7dc153c9f669b9a3b61a1664dd9cc05748818d559dcdf0448
meta/mlgym-bench
Community catalog
Technical details
Dataset reference
harbor:meta/mlgym-bench@sha256:4637b4f2a71602911c17071a66a039dac70a8b8ce2c582b0e114c9d3adf4b412
itmo-autods/maize-yield-case
Community catalog
Technical details
Maize grain-yield case (G2F G3 2023)
Dataset reference
harbor:itmo-autods/maize-yield-case@sha256:ca08ae3d755667fb17c4974df0b815ee0c78fac9ca6f73cf98f6ee7b19f3c78e
blobfishai/hubbench
Community catalog
Technical details
HubBench: one oracle-proven, deterministically graded Blobfish benchmark family per Harbor Hub professional-domain cluster — 104 stateful multi-system employee-decision tasks over isolated SQLite worlds (HubScore, no LLM judge)
Dataset reference
harbor:blobfishai/hubbench@sha256:5856ea4f83229e77b238a2acc3714c4d5e45329aed6e564819832a76cb33d5c6
billxbf/mimo-v26-general
Community catalog
Technical details
MiMo-V2.6-RL General: 925 Harbor tasks (converted from XiaomiMiMo/MiMo-V2.6-RL-oss)
Dataset reference
harbor:billxbf/mimo-v26-general@sha256:4ffb903ff0e4e473dcd3d20e76b37419e101986fe1e3c8365faec1dcaf12f673
doe/sample-pack
Community catalog
Technical details
Five Harbor tasks in legal, private equity and corporate card operations, graded by deterministic checks and a claude-opus-5-5 and gpt-6-astra judge panel
Dataset reference
harbor:doe/sample-pack@sha256:44760295e1ca08e7011b3992f2d1215c3420c3ae76d58b3f78772a551b8db8da
billxbf/mimo-v26-terminal
Community catalog
Technical details
MiMo-V2.6-RL Terminal: 64 Harbor tasks converted from XiaomiMiMo/MiMo-V2.6-RL-oss
Dataset reference
harbor:billxbf/mimo-v26-terminal@sha256:acc8c3b737ce8aaf1956021c3bbaa926da21790dbae6f866d69eac107c9f6223
open-design/open-design
Community catalog
Technical details
Open Design Benchmark — AI agent evaluation on design-asset generation tasks (landing pages, dashboards, mobile onboarding, PPT reports, text layouts, incremental patches, multi-format export)
Dataset reference
harbor:open-design/open-design@sha256:c6f811ac227a8667973ec588c10f03ab522f5438582a4de6b3d6f9dae7cbb8c1
blobfishai/erpbench-100-suite
Community catalog
Technical details
100 executable Oracle-Fusion-shaped ERP workflows with deterministic ERPScore grading
Dataset reference
harbor:blobfishai/erpbench-100-suite@sha256:26648e276ee6e190c73c343bf93cf77823718c9c565f7d92a330b4ddbe5fdc56

Next results