Agent Arena Index launches: benchmarks agents on ~3M real-world sessions
infwinston · x · 2026-11-19
Agent Arena Index is now live, built on nearly 3 million real-world agentic sessions to measure how well agents get actual work done and track the frontier of intelligence as it advances.
Alongside the main index, six new sub-indexes break down agent performance in depth:
- Knowledge work
- Coding
- Task completion
- Tool use
- Steerability
- And more
The thesis: the real world is the ultimate benchmark, in contrast to static exam-style evals.
Related event: Arena Launches Agent Arena Index Built on 2.9M Real Sessions(8 posts)→
More from Research
- 'Naturals + infinity' form a nonstandard model of Robinson arithmetic — jessi_cata · 2026-11-19
- Study: Claude investigator agents correctly identify the purpose of most alignment evals — CFGeek · 2026-11-19
- 7,000+ human gaze trials used to train gaze-guided visual reasoner GGR — LakeBrenden · 2026-11-19
- StreamEmbed: On-device real-time visual indexing of your screen history, 3.7x faster — CShorten30 · 2026-11-19
- Stanford's Anshul Kundaje disputes TF binding paper, cites ENCODE4 MPRA evidence — anshulkundaje · 2026-11-19
- Yacine revisits LeCun's 1988 paper on a theoretical framework for back-propagation — yacinelearning · 2026-11-19