AgentHorizon
Harbor: pendingA benchmark of 1,373 computer-use tasks with paired instructions to evaluate agentic judges on long-horizon trajectory success verification across three OSs.
- Samples
- 1,373
- Released
- 2026
- Scores
- 10
EfficientOpt
EfficientOpt: The Efficiency Gap in LLM-based Optimization Modeling
A benchmark of 561 expert-reviewed optimization tasks evaluating the correctness and computational efficiency (solver time) of LLM-generated mathematical models against paired ordinary and expert reference implementations.
How to run
ltb · efficientopt
$ npm i -g lettokenburn-cli$ ltb bench info efficientopt
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
11 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Qwen 3.5 122B | 69.0 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| GLM-5.1 | 70.0 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| GPT-5.5 | 72.1 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Claude Opus 4.6 | 73.0 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Qwen3-32B | 75.0 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Gemini 3.1 Pro | 79.1 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| MiniMax M2.5 | 83.0 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Kimi K2.6 | 84.5 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| DeepSeek V4 Flash | 85.3 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Qwen3.6 Plus | 91.4 | N/A | Independent | medium | Sep 30, 2026 | arxiv.org |
| Qwen3.6 27B | 92.1 | N/A | Independent | medium | Sep 30, 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
Content & Construction: EfficientOpt is built using OptTips, a knowledge base of 50 modeling techniques. The multi-agent framework OptDachshund transforms existing optimization problems into tasks that include natural-language descriptions, numerical data, and paired references (ordinary vs. expert) for cost comparison.
Scoring & Findings: Tasks are scored on correctness (optimal value match) and efficiency (solver runtime). Evaluation reveals an 'efficiency gap' where LLM-generated correct models often take significantly longer to solve than expert counterparts due to suboptimal modeling choices, despite achieving the same optimal values.