3DVerifierBench
Harbor adapterDVerifierBench Grounding Generative Verification in 3D ContentShuanglin Li, Jiaying Lin, Shuquan Ye.
- Samples
- 2,581
- Released
- 2026
- Scores
- 10
HICO-Bench
Hidden Information and Checkable Outcomes Benchmark
A benchmark of 180 paired everyday coordination scenarios evaluating social agents on plan feasibility and conversational integrity via verifiable outcomes.
How to run
ltb · hico-bench
$ npm i -g lettokenburn-cli$ ltb bench info hico-bench
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
12 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 80.8 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GPT-5.6 Sol | 76.3 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GLM-5.3 | 67.1 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Kimi K3 | 66.3 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| google/gemma-4-31B-it | 63.8 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Qwen3.8 Max | 60.8 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| DeepSeek V4 Pro | 59.2 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Qwen3.8 27B | 54.6 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GPT-5.6 Luna | 52.1 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GPT-4o | 51.3 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| MiniMax-M3 | 50.4 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Qwen3 8B | 20.4 | N/A | Independent | medium | Oct 14, 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
HICO-Bench comprises 180 scenarios across six coordination families where two agents with private information converse to reach a plan. Each scenario has two versions differing in one private condition, requiring distinct correct plans.
Scoring combines rule-based Feasibility (matching records satisfying private conditions) and LLM-judged Integrity. It finds 36% of highly-rated conversations fail verification, with best reported Verified Success at 28.3%.