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

1tests in this plan
1verified public harness
0need access or recipe work

Blocked tests stay in the plan. Public code alone does not reproduce an evaluator’s score.

2. Review the plan

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

  1. Read the linked maintainer guide and pin the harness commit.
  2. Configure the exact model adapter, dataset revision, split and task limit.
  3. Match prompts, tools, sampling, repetitions and judge to the intended run.
  4. 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 →