Open-source agent burns 106M tokens in 52 hours chasing an open math problem
GuiltyBookkeeper4849 · reddit · 2026-09-22
A developer is running an open-source autonomous agent against the open covering design problem C(25,15,5), where the best known solution uses 42 groups and the goal is 41.
- Live for 52+ hours, having processed 106,406,666 tokens; best solution so far improved to 44 groups
- All internal reasoning, memories and attempts are publicly inspectable (artificium-covering-experiment.gr.bio), with a real-time independent verifier for correctness
- Powered by open-source Qwen 27B plus a custom harness for autonomous work and continual learning (GitHub: officialgr/agent-artificium)
- Reproducible on consumer hardware (5090 on RunPod, 3090 feasible); reaching 41 would be the first open-source-agent math research breakthrough
- Author is seeking donations and compute sponsors
More from coding & agent
- User automates entire video clipping workflow with Nous Research's Hermes agent — Teknium · 2026-11-12
- As models improve, writing good evals gets harder and more important — amankhan · 2026-11-12
- Prime agent ships with day-one multi-agent communication, runs in Minecraft unmodified — sarahcat21 · 2026-11-12
- Autonomous agent edits just 1,024 weights to lift Qwen3.5 0.8B from 7.4% to 37.4% on MATH-500 — VictorSanh · 2026-11-12
- Omarchy 4.6 lets you share local AI models across your tailnet, says DHH — alexcovo_eth · 2026-11-12
- Coding harnesses hand shell access to inference providers — 4-step gateway mitigation — shensi · 2026-11-12