APEX-Harm
Harbor: pendingA benchmark evaluating prompt and script injection safety for long-horizon agents in shared professional workspaces across finance, consulting, and legal tasks.
- Samples
- 180
- Released
- 2026
- Scores
- 10
DrugTargetWorld
DrugTargetWorld: A Synthetic Biobank for Training and Benchmarking AI Scientists
A synthetic biobank benchmark evaluating AI agents on end-to-end drug target discovery across 540 episodes with known causal ground truth.
How to run
ltb · drugtargetworld
$ npm i -g lettokenburn-cli$ ltb bench info drugtargetworld
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Sonnet 5 | 21.3 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| GPT-5.6 Sol | 15.8 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Claude Opus 5 | 14.8 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Claude Haiku 4.5 | 12.9 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| gpt-oss-20b | 7.4 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| GLM-4-32B | 4.8 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Qwen3 8B | 4.6 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
| Devstral Small | 3.3 | N/A | Independent | medium | Oct 7, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
Procedurally generates simulated biobanks with genotypes, proteins, health records, and MRI data for 54,000 participants. Each world has a hidden causal model specifying disease drivers and biases.
Agents are scored on a 100-point composite metric covering target identification, causal confidence, and intervention direction. Top agents scored ~40/100, struggling to distinguish causal from non-causal proteins.