Capabilities · Concept
OpenDiscoveryTrace: Process Traces for AI Scientist Workflows
Public dataset of 558 full AI scientific-agent trajectories with step-level thoughts, tools, errors, and confidence across 124 science tasks.
Forecast horizon
Connections
Connections · 3
How this node ties into the rest of the map, and the evidence behind each link.
to · tracked by 4/10
OpenAI Dataset includes balanced GPT-5.4 trajectories among frontier models on scientific agent tasks.
+4 growth to · tracked by 4/10
Claude Fable 5 / Claude Mythos 5 / Claude Opus 4.8 (Anthropic) OpenDiscoveryTrace evaluates Claude Opus 4.6 trajectories on drug discovery, materials, genomics, and literature tasks.
+4 growth to · enables 3/10
AI Agents Reshaping Knowledge Work Process traces make AI scientist methodology auditable beyond final papers or code artifacts.
+3 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
Task horizon doubling Mar 2025
The length of software tasks AI agents can do autonomously at 50% reliability has doubled about every 7 months — and since 2024 closer to every ~3 months.
METR → Benchmarks saturating Apr 2025
In one year scores rose by 18.8, 48.9 and 67.3 points on MMMU, GPQA and SWE-bench; real-world software solve rate jumped from 4.4% to 71.7%.
Stanford HAI — AI Index 2025 → Autonomous coding 2025–2026
On SWE-bench Verified (500 real GitHub issues), autonomous coding agents reached ~80–86% by late 2025, up from under 50% in early 2025.
Epoch AI →