AI Agent Reliability Benchmark 2026: 65% Once, 25% Always
August 26, 2026
ThinkingBox is a new AI agent reliability benchmark: the best model scores 65.36% pass@1 but only 25.25% pass^20 across 507 stateful business agent tasks.
#agent-evaluation
Muse Code Benchmarks 2026: The Harness Changed Too
August 7, 2026
Meta's Muse Code posts a big Terminal-Bench 2.1 gain over Muse Spark 1.1 — but the harness changed too. What the verified leaderboard says a harness is worth.
AI Agent Reliability 2026: Structure Beats Instructions
August 5, 2026
Three July 2026 papers measured AI agent reliability: a verification loop added 1.5 points, guardrails recovered 19.9% of failures, prompt rules barely moved.
Claude Agent SDK Observability With OpenTelemetry (2026)
July 15, 2026
Instrument a Claude Agent SDK loop with OpenTelemetry in TypeScript: enable trace export, read the claude_code span tree, and ship real spans to a backend.