Docs / More

Per-task medians (2.x)

Per-task token and success medians from the Prism 2.x real-agent benchmark.

Per-task medians from the 2.x benchmark. Raw sessions: 2.x.json.

Per task (medians)

HostTaskA tokensB tokensA successB success
Claude Codees-toolkit-15043d8e240,748274,1712/22/2
Claude Codees-toolkit-2c4c0478292,002238,1232/22/2
Claude Codees-toolkit-323893f2279,058253,6012/22/2
Claude Codees-toolkit-3433044c713,050775,2612/22/2
Claude Codees-toolkit-827b92f7235,230310,2442/22/2
Claude Codees-toolkit-99af1860181,074145,5062/22/2
Claude Codees-toolkit-d9574efd284,307277,1552/22/2
Claude Codees-toolkit-e62aa15f401,004502,4112/22/2
Claude Codehono-afb2068c132,792138,8572/22/2
Claude Codehono-c3053ccf118,871121,7222/22/2
Claude Codehono-e5bb2062130,730157,6362/22/2
opencodees-toolkit-15043d8e912,158441,7570/11/1
opencodees-toolkit-2c4c0478506,939309,0931/10/1
opencodees-toolkit-323893f2575,588249,0871/10/1
opencodees-toolkit-3433044c1,391,7751,030,5870/11/1
opencodees-toolkit-827b92f7619,099414,1690/10/1
opencodees-toolkit-99af1860497,996240,1431/11/1
opencodees-toolkit-d9574efd574,282540,5461/11/1
opencodees-toolkit-e62aa15f330,705680,1451/11/1
opencodehono-afb2068c823,491547,3741/11/1
opencodehono-c3053ccf188,964347,2031/11/1
opencodehono-e5bb2062486,401413,4921/11/1

On this page