Unverified50% confidenceConceptExact time
SWE-Marathon stretches tasks into dozens or even hundreds of steps of continuous editing, testing the agent's ability not to get lost in a long context.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在代码Agent场景下，一次完整任务的轨迹长度可达数十万token73% similarUnverifiedHeavy-load agent tasks continuously occupy inference slots for seconds to tens of seconds, placing far greater pressure on GPU memory and KV cache than short conversations.71% similarUnverified一个复杂的 Agentic 任务可能需要数十次 LLM 推理，总耗时从几十秒到数分钟不等71% similarUnverifiedCompleting a full task (run) often requires multi-turn dialogue, tool calls, context retrieval, and other multi-step operations, and the cumulative token consumption can be tens of times higher than a single question-and-answer exchange.70% similarUnverifiedAutonomous machine learning engineering (MLE) tasks are inherently long-horizon endeavors, with the entire process often requiring dozens or even hundreds of consecutive decision steps.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1011217API
curl https://kongchang.com/api/v1/knowledge/claims/1011217MCP
get_claim(id=1011217)