3DVerifierBench
Harbor adapterDVerifierBench Grounding Generative Verification in 3D ContentShuanglin Li, Jiaying Lin, Shuquan Ye.
- Samples
- 2,581
- Released
- 2026
- Scores
- 10
TiGanEval
A benchmark of software engineering and research tasks measuring outcome alignment by evaluating agents on partial instructions where true requirements must be inferred from the environment.
How to run
ltb · tiganeval
$ npm i -g lettokenburn-cli$ ltb bench info tiganeval
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 | |||
|---|---|---|---|---|---|---|
| MiMo V2.6 Flash | 82.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 81.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| MiMo V2.6 Pro | 81.3 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GLM-5.3 Flash | 78.1 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Kimi K3 | 75.0 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.6-terra | 68.8 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.6 Sol | 64.1 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| GPT-5.6 Luna | 60.9 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3.6 35B A3B | 15.6 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| google/gemma-4-31B-it | 1.6 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| qwen3.5-9B | 0.0 | N/A | Independent | medium | Oct 13, 2026 | arxiv.org |
| Qwen3.8 27B | 0.0 | N/A | Independent | medium | Oct 13, 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
TiGanEval contains software engineering and research tasks where initial instructions are incomplete. True requirements are fixed and discoverable from code/runtime, separating stated from inferred needs.
Scoring uses success rate. Frontier models lose 53-67 points when moving from full to partial specs, indicating outcome alignment is broadly missing despite feedback or planning.