DRACO benchmark still puts fusion systems ahead of single frontier models
iamtrask · x · 2026-07-22
OpenMined reran an OpenRouter-style experiment on DRACO, a deep-research benchmark, and found that fusion models still dominate.
- The top 5 systems are all fusions, and the best single frontier model only placed 6th.
- The post says this matches observations from four startups and Los Alamos National Lab.
- The chart shows fusion combinations such as Fable + GPT-5.5, Opus + GPT-5.5 + DeepSeek, and DeepSeek + Kimi + GPT-5.5 outperforming solo models by a wide margin.
The author frames this as evidence that combining models remains the best path on deep-research tasks, and that more benchmarks are coming.
Related event: DRACO Benchmark Shows Routing Systems Outperform Single Frontier Models(2 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27
- Meta FAIR's Self-Direction turns agent failures into reusable LoRA skills, +13.8pp on Terminal-Bench — karen_ullrich · 2026-10-27