Read original ↗

paperarXivTrust 82 · PrimaryPublished 1mo agoLive · 1mo ago

Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent

We introduce Agents-A1, a 35B Mixture-of-Experts Agentic Model that reaches trillion-parameter-level performance by scaling the agent horizon. We investigate agent-horizon scaling from two perspectives: scaling long-horizon trajectories and scaling heterogeneous agent abilities. To support this goal, we build a long-horizon knowledge-action infrastructure that connects external knowledge, actions, observations, and verifier outcomes, producing agentic trajectories with an average length of 45K tokens. Based on this, we train Agents-A1 with a three-stage recipe. First, we perform full-domain su

Lineage graph

Paper → model → repo connections mined from source citations (Tier-1 exact match).

Why these links exist

Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.

LinkedLinked via unknownGeneral Intuition’s $2.3B bet that video games can train AI agents for the real world →
LinkedLinked via unknownAlibaba's model never trained as an agent — and improved agent performance across seven benchmarks →
LinkedLinked via unknownGoogle DeepMind is worried about what happens when millions of agents start to interact →
LinkedLinked via unknownMonitor and debug generative AI inference with SageMaker detailed metrics and Insights dashboard on CloudWatch →
LinkedLinked via unknownFabiojvv/ai-cortex-hub →
FuzzySimilar title/name (fuzzy) · 59%AgentCore-8B →
“Fuzzy title match (0.73): “Scaling the Horizon, Not the Parameters: Reaching Trillion-P” ≈ “AgentCore-8B””
LinkedLinked via unknownDeepSeek open sources DSpark, a new framework to speed up LLM inference by up to 85% →
LinkedLinked via unknownClaude Meets Blackwell Ultra: Anthropic’s Models Now Run on NVIDIA GB300 in Azure →

Covers

newsGeneral Intuition’s $2.3B bet that video games can train AI agents for the real world newsAlibaba's model never trained as an agent — and improved agent performance across seven benchmarks newsGoogle DeepMind is worried about what happens when millions of agents start to interact newsMonitor and debug generative AI inference with SageMaker detailed metrics and Insights dashboard on CloudWatch

Implements

repoFabiojvv/ai-cortex-hub

Has model

modelAgentCore-8B

Covers (incoming)

newsDeepSeek open sources DSpark, a new framework to speed up LLM inference by up to 85%newsClaude Meets Blackwell Ultra: Anthropic’s Models Now Run on NVIDIA GB300 in Azure newsInto the Omniverse: Three Workflows for Improving Vision AI Agent Accuracy With Synthetic Data and Fine-Tuning newsNVIDIA Vera Rubin Maximizes Intelligence per Dollar for Post-Training Workloads – a Key Metric for Agentic AI

Implements (incoming)

reporeunios2024/cortex-sentinel-trading-nexus repotensorflow/serving repodeepset-ai/haystack-core-integrations repodeepset-ai/haystack-integrations repoSWE-agent/SWE-agent repozhayujie/CowAgent repoThysrael/Horizon repoNousResearch/hermes-agent repoiflytek/astron-agent

Related across the graph

repoSWE-agent/SWE-agent newsInto the Omniverse: Three Workflows for Improving Vision AI Agent Accuracy With Synthetic Data and Fine-Tuning newsDeepSeek open sources DSpark, a new framework to speed up LLM inference by up to 85%newsGeneral Intuition’s $2.3B bet that video games can train AI agents for the real world repotensorflow/serving newsClaude Meets Blackwell Ultra: Anthropic’s Models Now Run on NVIDIA GB300 in Azure repoFabiojvv/ai-cortex-hub newsMonitor and debug generative AI inference with SageMaker detailed metrics and Insights dashboard on CloudWatch newsNVIDIA Vera Rubin Maximizes Intelligence per Dollar for Post-Training Workloads – a Key Metric for Agentic AI repozhayujie/CowAgent repoNousResearch/hermes-agent repodeepset-ai/haystack-integrations newsGoogle DeepMind is worried about what happens when millions of agents start to interact repodeepset-ai/haystack-core-integrations repoThysrael/Horizon newsAlibaba's model never trained as an agent — and improved agent performance across seven benchmarks modelAgentCore-8B repoiflytek/astron-agent reporeunios2024/cortex-sentinel-trading-nexus

Topics