Apodex launches TRACES, a benchmark grading AI on open-ended discovery instead of known answers
Faheem_uh · x · 2026-09-11
Apodex (founded by Tianqiao Chen) launched TRACES, a new benchmark paradigm for "Discoverative AI" that measures whether AI can find answers that don't yet exist, with a live leaderboard.
- Setup: 10 STEM PhDs spent two months scouting 561 industries across 16 sectors, building a registry of 423 real-world problems; 20 are live as executable environments (AAV capsid design, drug repurposing, clinical trials, LLM engineering). Current benchmark: 1,182 trajectories across 17 environments.
- Scoring: The HDS6 rubric grades the process, independent of the final answer — Tools, Repair, Alternatives, Coherence, Evidence, Scope.
- Early results (Apodex-attributed): an autonomously-trained specialist model hit human-expert level on rare-disease diagnosis (research result, not clinical), and a biomedical environment lifted drug-repurposing scores over the closed-book base model.
Related event: Apodex unveils TRACES benchmark for discovery-oriented AI(3 posts)→
More from Research
- HABIT-Bench: Best agent memory system scores just 33.61% at inferring user habits — SJTU · 2026-10-28
- MatAnyone-E brings 7.4x faster 4K video matting via sparse memory propagation — mmlab-ntu · 2026-10-28
- CodeCap-Bench: executable code beats natural language as video captions — Jiafu Tang · 2026-10-28
- NVIDIA's ThinkWorld learns robot skills from unlabeled human videos via chain-of-dynamics reasoning — nvidia · 2026-10-28
- Omni-InteractScope benchmarks proactive interaction in continuous omnimodal streams — Qwen · 2026-10-28
- Memory Tax study: dedicated agent memory systems often lose to naive RAG — UIUC-CS · 2026-10-28