LLM Digest

Story

arxiv_cs_ai · Sep 2, 2026 · paper

Source brief

CivBench: A Long-Horizon Benchmark for Tool-Mediated Agents in Civilization VI

arxiv.orgSep 2, 2026
original source linked

In brief

We present CivBench, an open-source benchmark for evaluating language model agents in long-horizon, tool-mediated environments through the Model Context Protocol (MCP). A single episode spans 300+ turns and produces t...

Feed lens

agenteval

Read the original at arxiv.org →Open in live feed Read that day’s brief

CivBench: A Long-Horizon Benchmark for Tool-Mediated Agents in Civilization VI

Earlier in this thread 4 items

Parsing the Stream: A Live Trace Model for Long-Horizon Agents and Their Observers

PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning

Emergence World: A Laboratory for Evaluating Long-Horizon Agent Autonomy

Governance Decay: How Context Compaction Silently Erases Safety Constraints in Long-Horizon LLM Agents