Papers
Browse the full research library.
Ten papers per page, with summaries, diagrams, formulas, concepts and citations kept close to the source.
Muse Spark Open
A new open-weight multimodal model for coding agents: 816B total parameters, 42B active, 1M context, multiple inference runtimes and BF16/NVFP4 checkpoints.
Read this for a concise, technically grounded view of Muse Spark Open’s architecture, modalities, hardware footprint, inference options, license and limits. Benchmarks and vendor claims are explicitly attributed rather than treated as independent results.
Meta Superintelligence Labs
Read paper
Harness²
Eight editable harness components, two recursion loops, benchmark commands, code requirements, full reported score deltas, and production safety caveats.
Read this if you build production agents and want to improve reliability without fine-tuning a new model. The paper explains both the refinement loop and the engineering caveats around evaluation, cost, and benchmark leakage.
Ruiyao Xu, Yanfei Chen, Zhongying CuiZhu
Read paper
Decisions API vs Jev
Compare two machine-oriented AI decision APIs, their output contracts, input support, and practical trade-offs.
Useful for AI engineers designing classification, routing, scoring, and automation workflows.
OpenAI, TypeSafe AI
Read paper
OpenAI Decisions API
OpenAI has introduced a dedicated Decisions API for fast, bounded AI decisions. It accepts shared text or multimodal evidence and lets developers ask predicate, choice, or score questions whose typed answers can feed application logic directly.
Read this if you build AI agents, routing systems, classifiers, guardrails, multimodal workflows, or production AI infrastructure and want to understand where a decision-specific API fits alongside Responses, Structured Outputs, and function calling.
OpenAI
Read paper
Evaluation and Benchmarking of LLM Agents: A Survey
Evaluation and Benchmarking of LLM Agents: A Survey: Reliability, safety and capability evaluation.
Read this paper to understand reliability, safety and capability evaluation in LLM-agent design.
Mahmoud Mohammadi
Read paper
SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering
SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering: Coding agents and agent-computer interfaces.
Read this paper to understand coding agents and agent-computer interfaces in LLM-agent design.
John Yang
Read paper
OpenHands: An Open Platform for AI Software Developers as Generalist Agents
OpenHands: An Open Platform for AI Software Developers as Generalist Agents: Production-oriented coding agents.
Read this paper to understand production-oriented coding agents in LLM-agent design.
Xingyao Wang
Read paper
Cognitive Architectures for Language Agents
Cognitive Architectures for Language Agents: Memory, planning and action architecture.
Read this paper to understand memory, planning and action architecture in LLM-agent design.
Theodore R. Sumers
Read paper
The Rise and Potential of Large Language Model Based Agents: A Survey
The Rise and Potential of Large Language Model Based Agents: A Survey: Complete agent landscape.
Read this paper to understand complete agent landscape in LLM-agent design.
Zhiheng Xi
Read paper
WebArena: A Realistic Web Environment for Building Autonomous Agents
WebArena: A Realistic Web Environment for Building Autonomous Agents: Web and computer-use agents.
Read this paper to understand web and computer-use agents in LLM-agent design.
Shuyan Zhou
Read paper