AgenticBBO-Bench
Harbor: pendingA cross-domain benchmark evaluating LLM agents on five black-box optimization tasks (synthetic, HPO, DB tuning, chip design, molecular) under a unified finite-budget protocol.
- Samples
- N/A
- Released
- 2026
- Scores
- 7
AAArena
Agents for Agents Arena
A benchmark comprising 12 authentic adversarial games and 1,920 archived human programs to evaluate AI agents' ability to iteratively refine executable game policies via heuristic learning.
How to run
ltb · aaarena
$ npm i -g lettokenburn-cli$ ltb bench info aaarena
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
7 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Opus5.5 | 2703.9 | claude code | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-6 Sol | 2418.2 | codex | Independent | medium | Oct 8, 2026 | arxiv.org |
| GLM-5.3 | 2359.2 | codex | Independent | medium | Oct 8, 2026 | arxiv.org |
| DeepSeek V4 Pro | 2288.2 | codex | Independent | medium | Oct 8, 2026 | arxiv.org |
| Kimi K3 | 2288.2 | codex | Independent | medium | Oct 8, 2026 | arxiv.org |
| Qwen3.8 Max | 2288.2 | codex | Independent | medium | Oct 8, 2026 | arxiv.org |
| LongCat 2.0 | 2069.9 | codex | Independent | medium | Oct 8, 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
AAArena evaluates Adversarial Heuristic Learning by tasking AI agents with interpreting rules, selecting opponents, and revising code policies across 12 games using a frozen pool of 1,920 human programs.
Performance is measured by rank and Elo against human ladders within fixed match budgets. The strongest configuration achieved rank-1 in 6 of 12 games, highlighting challenges in complex rule understanding.