Time per task: agentic coding session
One long tool-using loop: 60,000 tokens in across turns, 4,000 tokens out. Time = time to first answer + output tokens ÷ output speed.
As of 15 Nov 2026, BenchLeader ranks Muse Ember first on the Time per task: agentic coding session board at 13.6 s, ahead of Gemini 2.5 Flash-Lite (15.9 s) and Gemini 3.1 Flash Lite (16.8 s).
461 models · data as of 15 Nov 2026
- 1Muse Ember (xhigh)13.6 s
- 2Gemini 2.5 Flash-Lite (none)15.9 s
- 3Gemini 3.1 Flash Lite16.8 s
- 4Gemini 3.5 Flash Lite18.4 s
- 5Mercury 218.4 s
- 6Beam 1.0 501B A23B19.3 s
- 7DeepSeek V4.1 Flash (none)19.6 s
- 8Nova Micro20.0 s
- 9Beam 1.021.3 s
- 10Muse Spark 1.122.5 s
- 11Gemini 2.5 Flash (none)23.4 s
- 12Nemotron 3 Ultra 550B A55B (thinking)24.6 s
- 13GPT-4.1 nano25.2 s
- 14gpt-oss-120b (low)25.3 s
- 15Gemini 3 Flash (thinking)25.5 s
219 of 219
Cite as: BenchLeader, “Time per task: agentic coding session”, data as of 15 Nov 2026, https://www.benchleader.com/leaderboards/time-agent-session.