Context · Frontier AI · sweep Tuesday, November 17, 2026
Code agent benchmarks for long-horizon software construction
The evaluation of programming agents migrates from generating isolated files to engineering repositories and complete products with persistent state, which redefines the standard of commercial viability for agents. The bottleneck is no longer model capability but the measurement of continuity, maintenance, and delivery in real environments (Replit, Unity, Roblox), where compute infrastructure and inference cost determine which agents survive.
Title, thesis and gravity assessment written by an AI model, without prior human review.
Gravity35/100 · Background
Why this gravity
Los 12 artículos son preprints de arXiv que proponen benchmarks (ViBench 2.0, GenXR-Bench, UnityBench-Mini3D, OpenGameEval, CEGame-Bench) para agentes de código en horizontes largos; Replit, Unity y Roblox aparecen como entornos de evaluación, no como anuncios corporativos ni adopción industrial declarada, y no hay métricas de discontinuity ni release de modelo frontier.
- Coverage
- 1 feed · 1 independent publisher group (a single publisher group)
- Cross verticals
- Frontier AI, Tech Industry, Startups
- Tier 1 actors
- OpenAI
- Gravity vs. the previous edition
- +3 (32 → 35)
coverage, not accuracy
Lead source
GenXR-Bench: Benchmarking Coding Agents for Interactive XR Development · arXiv CS.AI / William Huang, David Li, Xiang 'Anthony' Chen, Xun Qian, Nels Numan, Zhongyi Zhou, Hyunsung Cho, Salman Muin Kayser Chishti, Ding Xia, Xincheng Huang, Jiahao Ren, Faraz Faruqi, Tianyu Xu, Yang Zhang, David Kim, Ruofei DuTimeline · Madrid time
- 06:00
- 06:00
- 06:00
- 06:00FDE-Gym: The Last Mile of AI Delivery · arXiv CS.AI
- 06:00
- 06:00
- 06:00
- 05:00
- 05:00
- 05:00
- 06:00
- 06:00
ShareWhatsApp
Related stories
- 0Persistent memory in vision-language-action models for robotic navigation and manipulation
- 80Legal liability for uncontrolled AI agents at OpenAI
- 80DeepSeek expansion in agents, APIs and inference infrastructure
- 80Moonshot's Kimi K3.1 Pro and open weights in frontier models
- 80Automation of vulnerability discovery and exploitation by AI