ARC-AGI-2
Harbor nativeARC-AGI-2: 更难的抽象推理
Verified100%Abstract Reasoning
- Samples
- 167
- Released
- 2025
- Scores
- 118
PLCWorld
PLCWorld: Benchmarking LLM-Generated PLC Programs in Closed-Loop Plant Simulation
A benchmark with 100 synthetic tasks and 473 task-condition pairs evaluating LLM-generated Structured Text PLC programs via closed-loop plant simulation for task success and safety.
How to run
ltb · plcworld
$ npm i -g lettokenburn-cli$ ltb bench info plcworld
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
6 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-4o | 92.8 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Claude Sonnet 5 | 50.9 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Qwen3.5 Plus | 49.8 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| GPT-5.5 | 37.9 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Gemini 3.1 Pro | 36.2 | N/A | Independent | medium | Oct 2, 2026 | arxiv.org |
| Gemini 3.8 Flash | 32.1 | N/A | Independent | medium | Oct 2, 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
PLCWorld contains 100 synthetic industrial control tasks (Motion Control, Material Handling) with 473 registered task-condition pairs. It couples Structured Text execution with simulated plant responses to assess control dependencies.
The protocol reports Task Success and Safety Violation separately. Direct GPT-5.5 achieved 82.70% success on Easy cases but only 25.10% on Hard cases, highlighting significant execution gaps.