AI
Opera: A Verbal Critic Framework for Long-horizon Coding Agents
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 07/10/2026, 20:00:00 UTC
Briefing created: 20/11/2026, 03:15:49 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 21:18:24 UTC.
Briefings
Opera is a verbal critic framework for long-horizon coding agents that treats corrections as persistent notes until problems are resolved. It improves resolve rates by up to 12.4 percentage points on Terminal-Bench 2.1 and 15.0 on SWE-Bench Pro subsets across four policy models. Opera also provides on-policy training data via rollouts, boosting fine-tuning performance by 10.2 percentage points without inference-time criticism.
Key points
- Opera treats each correction as a persistent note that remains active until the diagnosed problem is resolved, unlike traditional feedback that may be ineffective or harmful.
- Opera improves policy model resolve rates by up to 12.4 percentage points on Terminal-Bench 2.1 and 15.0 on SWE-Bench Pro subsets across four different policy models.
Context
Long-horizon coding agents require timely corrections, but existing critics often fail to track post-feedback actions. Opera addresses this by treating corrections as persistent notes until resolution, improving resolve rates on SWE-Bench Pro and Terminal-Bench 2.1 benchmarks.
Limitations
Opera's performance data is limited to specific benchmark subsets and policy models; results may vary across different environments or agent configurations not tested in the study.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 20/11/2026, 03:15:49 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeibziscil46zyx56hkvdyq55ohlevn4ezhkuomztztsbz5y42d53ba