OverviewReproduce a result
Reproduce a result
Build a full local run plan with every test, recipe and missing setting kept explicit.
1. Choose the model and tests
Blocked tests stay in the plan. Public code alone does not reproduce an evaluator’s score.
2. Review the plan
τ³-BankingEvaluator access
A complete executable recipe is not catalogued here. An evaluator result is not proof that its private tasks or judging pipeline can be run locally.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Read the evaluator’s methodology and access terms.
- Request the matching evaluation version and complete run configuration.
- Obtain task, prompt, judge and submission access.
- Record the evaluator’s run receipt and result artifact.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →IFBenchEvaluator access
A complete executable recipe is not catalogued here. An evaluator result is not proof that its private tasks or judging pipeline can be run locally.
Pinned dataset: allenai/IFBench_test ↗ @ ed7c12452594 · train · Benchmark author · Open
- Read the evaluator’s methodology and access terms.
- Request the matching evaluation version and complete run configuration.
- Obtain task, prompt, judge and submission access.
- Record the evaluator’s run receipt and result artifact.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →SWE-Bench Pro V2 HARDEvaluator access
A complete executable recipe is not catalogued here. An evaluator result is not proof that its private tasks or judging pipeline can be run locally.
Pinned dataset: ScaleAI/SWE-bench_Pro ↗ @ 2d52cb3df914 · test · Benchmark author · Open
- Read the evaluator’s methodology and access terms.
- Request the matching evaluation version and complete run configuration.
- Obtain task, prompt, judge and submission access.
- Record the evaluator’s run receipt and result artifact.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
- Licence review for hosted or redistributed use
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →SWE-Bench Pro v1Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: ScaleAI/SWE-bench_Pro ↗ @ 2d52cb3df914 · test · Benchmark author · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
- Licence review for hosted or redistributed use
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →SWE Atlas Codebase QnARecipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: ArtificialAnalysis/SWE-Atlas-QnA-AA ↗ @ 1f28d51b6e19 · test · Evaluator adaptation · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →SWE-bench MultilingualRecipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: SWE-bench/SWE-bench_Multilingual ↗ @ 846e647b9f33 · test · Benchmark author · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →AIME 2026Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: math-ai/aime26 ↗ @ 79037aebdb65 · test · Benchmark author · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →AutomationBench (public split)Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: UpstageShareSpace/automationBench ↗ @ 4b68a3021302 · train · Community mirror · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →MCP AtlasRecipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: ScaleAI/MCP-Atlas ↗ @ 8c563b55d7c9 · train · Benchmark author · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →BrowseComp (context management)Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →DeepSearchQA (context management)Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →AA-LCR (version not specified)Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →LongBench v2Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
Pinned dataset: zai-org/LongBench-v2 ↗ @ 2b48e494f2c7 · train · Benchmark author · Open
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →AA-Omniscience (public split)Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →SciCodeEvaluator access
A complete executable recipe is not catalogued here. An evaluator result is not proof that its private tasks or judging pipeline can be run locally.
Pinned dataset: SciCode1/SciCode ↗ @ 4510f6a6aa27 · test · Benchmark author · Open
- Read the evaluator’s methodology and access terms.
- Request the matching evaluation version and complete run configuration.
- Obtain task, prompt, judge and submission access.
- Record the evaluator’s run receipt and result artifact.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →CritPtEvaluator access
A complete executable recipe is not catalogued here. An evaluator result is not proof that its private tasks or judging pipeline can be run locally.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Read the evaluator’s methodology and access terms.
- Request the matching evaluation version and complete run configuration.
- Obtain task, prompt, judge and submission access.
- Record the evaluator’s run receipt and result artifact.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →Terminal-Bench 2.1Evaluator access
A complete executable recipe is not catalogued here. An evaluator result is not proof that its private tasks or judging pipeline can be run locally.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Read the evaluator’s methodology and access terms.
- Request the matching evaluation version and complete run configuration.
- Obtain task, prompt, judge and submission access.
- Record the evaluator’s run receipt and result artifact.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →SWE-bench (Verified)Public harness
Maintainer instructions checked 24 October 2026. This command grades existing patches in Docker. An agent must first generate predictions with the chosen model; grading alone does not rerun model inference. Pin the agent, issue subset and Docker images. This is not a reconstruction of Vals’ bash-only harness. Inference, judging or local compute may incur costs; configure provider limits first.
Pinned dataset: princeton-nlp/SWE-bench_Verified ↗ @ c104f840cc67 · test · Benchmark author · Open
- Read the linked maintainer guide and pin the harness commit.
- Configure the exact model adapter, dataset revision, split and task limit.
- Match prompts, tools, sampling, repetitions and judge to the intended run.
- Run generation and scoring stages; retain raw outputs, task outcomes and logs.
swebench eval verified -p <model-predictions.jsonl> --run-id <unique-run-id> -j 1Author / evaluator instructions ↗1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
- Licence review for hosted or redistributed use
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →GPQA DiamondPublic harness
Maintainer instructions checked 24 October 2026. Use the Diamond task, not a different GPQA subset. Confirm the task ID with lm-eval ls tasks and pin the dataset and harness. The public harness is not the Vals or AA pipeline. Inference, judging or local compute may incur costs; configure provider limits first.
Pinned dataset: Idavidrein/gpqa ↗ @ 83022cefff93 · gpqa_diamond · Benchmark author · Gated
- Read the linked maintainer guide and pin the harness commit.
- Configure the exact model adapter, dataset revision, split and task limit.
- Match prompts, tools, sampling, repetitions and judge to the intended run.
- Run generation and scoring stages; retain raw outputs, task outcomes and logs.
lm-eval run --model <backend> --model_args <provider-configuration> --tasks <verified-gpqa-diamond-task> --output_path results --log_samplesAuthor / evaluator instructions ↗1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
- Dataset access approval for the running account
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →DeepSWE v1.1Recipe not catalogued
Public execution instructions have not been verified in this catalogue. This status does not claim that the benchmark is closed or impossible to run.
No dataset revision is pinned for this test. The plan keeps the missing dataset, harness and settings visible.
- Open the benchmark’s author page.
- Confirm code and dataset availability, licence, task split and scoring.
- Record the exact harness commit, model, sampling and judge before adding a command.
1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Dataset revision and split
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
A fresh evaluation cannot reproduce the published score exactly until its model version, data, harness, prompts, tools, judge and sampling settings are matched. Unknown source settings remain unknown.
Understand this test →Humanity's Last ExamPublic harness
Maintainer instructions checked 24 October 2026. Run from hle_eval. The separate run_judge_results.py stage is required to obtain a score. Pin judge model, prompts, token budget and full versus text-only split. Public HLE and HLE-Rolling are different datasets; this recipe covers public HLE only. Inference, judging or local compute may incur costs; configure provider limits first.
Pinned dataset: cais/hle ↗ @ 5a81a4c7271a · test · Benchmark author · Gated
- Read the linked maintainer guide and pin the harness commit.
- Configure the exact model adapter, dataset revision, split and task limit.
- Match prompts, tools, sampling, repetitions and judge to the intended run.
- Run generation and scoring stages; retain raw outputs, task outcomes and logs.
python run_model_predictions.py --dataset cais/hle --model <provider-model> --max_completion_tokens <budget> --num_workers 1 --max_samples <task-limit>Author / evaluator instructions ↗1 source-specific reference results for this model; each remains in the downloaded plan.
Configuration still required
- Immutable model or endpoint version
- Adapter verification against the pinned dataset
- Harness commit
- Prompt and tool policy
- Reasoning effort
- Sampling and seed
- Task limit and repetitions
- Judge and aggregation
- Provider spending limits
- Dataset access approval for the running account
The dataset revision and split are pinned in this job. Model version, harness, prompts, tools, judge and sampling remain unmatched, so a fresh run is not an exact reproduction claim.
Understand this test →