27B agent Faraday beats Claude Opus 4.8 and GPT-5.5 on research replication via new Replica method
omarsar0 · x · 2026-08-15
A new study introduces Replica, a method that turns paper replication into a scalable RL task space, and trains a 27B-parameter agent called Faraday that outperforms Claude Opus 4.8 and GPT-5.5 on held-out research replication tasks.
- Core idea: Replicating papers forces hypothesis-driven exploration similar to open research and surfaces underspecified details.
- Reward signal: An auto-generated rubric judge provides low-noise rewards aligned with human assessment.
- Technical highlights: Faraday uses coding agents as tools; rollout analysis shows it takes a more scientifically principled approach rather than gaming the rubric.
- Implication: Authors argue this points toward training long-horizon scientific capabilities into weights without complex harnesses.
Paper link included.
More from coding & agent
- Dispatch launches newsletter infrastructure built for AI agents, one API call to publish — RichardsonDx · 2026-10-16
- Creator shares the agent-powered system behind millions of monthly impressions — VibeMarketer_ · 2026-10-16
- Devs say LLM coding mistakes are a different class: overcomplication, inferred limits, worse at higher reasoning — Mu5_ · 2026-10-16
- How do you debug bad agent runs when you already have tracing? — Sensitive-Parsnip-12 · 2026-10-16
- Where does pre-task context live for agents? Devs debate workflows beyond handoff — shishen2026 · 2026-10-16
- No-instruction autonomous agent wakes up, sets own goals, and tries hacking by default — sethlazar · 2026-10-16