3DVerifierBench
Harbor adapterDVerifierBench Grounding Generative Verification in 3D ContentShuanglin Li, Jiaying Lin, Shuquan Ye.
- Samples
- 2,581
- Released
- 2026
- Scores
- 10
Realtime-Bench
Realtime-Bench: Can Frontier Models Solve Real-time Interactive Problems?
A suite of 50 dynamic interactive game and web tasks evaluating agents' ability to perceive evolving environments and act under strict temporal constraints using Wall, Paused, and Token clocks.
How to run
ltb · realtime-bench
$ npm i -g lettokenburn-cli$ ltb bench info realtime-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
11 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 42.7 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemini 3.8 Flash | 27.3 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| GPT-5.6 Sol | 18.0 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Claude Fable 5.1 | 16.0 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Claude Opus 5 | 15.3 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| GPT Realtime 2 | 15.3 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemini 3.1 Flash Live | 13.3 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Kimi K3 | 13.3 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Jev 1.13 | 12.0 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 8.7 | N/A | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemini 3.8 Live | 6.0 | N/A | Independent | medium | Oct 15, 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
Realtime-Bench comprises 50 carefully curated dynamic tasks in interactive games and web environments where delayed actions directly affect success. It evaluates real-time streaming capabilities across temporal state awareness, adaptive deliberation, and time-aware action selection.
Scoring uses Success Rate and Pass^3 (consistency across three trials) under Wall Clock, Paused Clock, and Token Clock protocols. The best reported success rate is 44%, indicating that real-time interaction remains a significant challenge for frontier models.