ARC-AGI-2
Harbor nativeARC-AGI-2: 更难的抽象推理
Verified100%Abstract Reasoning
- Samples
- 167
- Released
- 2025
- Scores
- 118
GameCommBench
GameCommBench: A Unified Benchmark and Type-Aware Evaluation for AI-Generated Game Commentary
A unified benchmark for AI-generated game commentary across board games, sports, and esports, featuring type-annotated data and TACE evaluation framework.
How to run
ltb · gamecommbench
$ npm i -g lettokenburn-cli$ ltb bench info gamecommbench
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-5.4 | 0.6 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Sonnet 4.5 (2025-09-29) | 0.5 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Gemini 3.1 Pro | 0.5 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Qwen3-VL-32B | 0.5 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| InternVL3.5-38B | 0.4 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GLM-4.6V | 0.3 | N/A | 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
GameCommBench spans chess, Go, soccer, basketball, and LoL, providing aligned game contexts and commentary annotated as descriptive, analytical, or background. It uses the Type-Aware Commentary Evaluation (TACE) framework to assess functional heterogeneity rather than holistic scores.
The benchmark reveals non-uniform capabilities in AI commentators, identifying live observation and strategic analysis as key bottlenecks. TACE is validated for reliability and human agreement, offering a diagnostic foundation for interpretable evaluation.