One Layer Deeper launches an H100-only competition for deeper reasoning
aryaman2020 · x · 2026-08-04
Core Automation and Tilde Research have launched One Layer Deeper, a competition that asks participants to make a model “think one layer deeper” under a fixed budget.
- Constraint: one file, one H100, and one fixed model-state ceiling.
- Goal: spend the training-time budget however you want, then submit a single final score.
- Current leaderboard: 16 ranked entries are already up.
- The rules say ranking is based on unrounded accuracy, with the target advancing from T=1 to 2, 4, 8, 16, 32, 64 as examples are solved.
- The benchmark tracks both in-distribution and out-of-distribution number-size performance.
Related event: Tilde Launches 'One Layer Deeper' Compute Competition(3 posts)→
More from Research
- Paper maps 41 agent failure modes to model, harness, tools and environment — omarsar0 · 2026-08-04
- Claude thread proposes an adversarial review skill for agentic science — ChrisGPotts · 2026-08-04
- Singular learning theory talk links neural-network generalization to alignment — andrewgwils · 2026-08-04
- AI paper argues best-of-K boosts generative expressivity, not just sampling quality — anshulkundaje · 2026-08-04
- ASCII art may be a better taste benchmark for frontier models than you think — weswinder · 2026-08-04
- A curated reading list for DeltaNet, FlashKDA, vLLM serving and MoE — austinvhuang · 2026-08-04