LetTokenBurnA benchmark of 110 tasks and over 1.1K executable criteria for evaluating AI agents' ability to edit executable worlds in Minecraft and Terraria.
$ npm i -g lettokenburn-cli$ ltb bench info igmbench
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-5.6 Sol | 78.2 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Claude Opus 4.8 | 71.8 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| GPT-5.6 Luna | 64.5 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Kimi K3 | 60.0 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| Gemini-3.5-flash | 57.3 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| GLM-5.3 | 48.2 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
| DeepSeek V4 Pro | 31.8 | N/A | Independent | medium | Oct 1, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
IGMBench comprises 110 world-editing tasks across Minecraft and Terraria, spanning property, entity, dynamics, and system interventions. It features over 1.1K executable state and behavioral evaluation criteria with deterministic validation.
Scoring involves checking executability, behavioral correctness, preservation, and visual consistency. Frontier agents solve 78.2% of tasks under strict criteria, with performance decreasing as intervention depth increases.