Lowest Latency AI Models: Minimized TTFT

Ranked by Time to First Token (TTFT). Best for low-latency interactive applications, chat interfaces, and high-performance responses.

Elite Performers

Frontier Model	LMArena Elo	API Cost (1M)	Throughput
nvidia-llama-3.3-nemotron-super-49b-v1.5	1343	$1e-7	69
qwen3-30b-a3b	1327	$8e-8	98
llama-3.1-nemotron-70b-instruct	1298	$0.0000012	12
llama-3.2-vision-11b-instruct	992	$2.45e-7	39
olmo-3.1-32b-instruct	1331	$2e-7	57
qwen2.5-coder-32b-instruct	1270	$6.6e-7	30
gemma-3-12b-it	1341	$4e-8	9

*These models represent the Pareto Frontier (optimal cost-to-performance).*