LetTokenBurnMine Odyssey: Benchmarking Spatial Agentic Intelligence in the Wild
A benchmark evaluating agentic spatial intelligence via 180 navigation tasks across 30 Minecraft reconstructions of real-world locations.
$ npm i -g lettokenburn-cli$ ltb bench info mine-odyssey
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
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 85.6 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Opus 5.5 | 73.9 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Claude Opus 5 | 50.6 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GPT-5.6 Sol | 43.9 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Gemini 3.8 Flash | 26.7 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| DeepSeek V4.1 Flash | 23.9 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| Grok-4.6 | 17.2 | N/A | Independent | medium | Oct 8, 2026 | arxiv.org |
| GLM-5.3 Flash | 15.6 | 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.
Comprises 180 tasks across 30 Minecraft maps of real-world sites (20 outdoor, 10 indoor) in 20 countries. Tasks require agents to follow natural-language instructions to visit ordered waypoints.
Success is measured by task completion rate. GPT-6 Astra leads with 85.6% success, while the best open-weight model reaches only 23.9%, indicating significant room for improvement.