llm.ing

Median time to first token

independentrun by Artificial Analysisunit: secondsDisplay with attribution; no redistribution47 modelssource ↗
AA Intelligence IndexAA Coding IndexAA Agentic IndexMedian output speedMedian time to first token
#ModelScoreObserved
1Gemini 2.5 Flash-Lite
0.29s
Sep 21, 2026
2Grok Build 0.1
0.52s
Sep 20, 2026
3GPT-5.4non-reasoning
0.80s
Sep 21, 2026
4MiniMax-M3
0.86s
Sep 21, 2026
5nemotron-3-nano-30b-a3b
0.94s
Sep 21, 2026
6GPT-4.1
0.96s
Sep 21, 2026
7Claude Opus 4.7non-reasoning
1.03s
Sep 20, 2026
8Claude Opus 4.5 (latest)
1.21s
Sep 20, 2026
9Claude Sonnet 4.6
1.31s
Sep 21, 2026
10GLM-5
1.51s
Sep 21, 2026
11MiniMax-M2.7
1.59s
Sep 2, 2026
12MiniMax-M2.5
1.62s
Sep 21, 2026
13GLM-4.7-Flash
1.67s
Sep 21, 2026
14MiniMax-M2.1
1.73s
Sep 21, 2026
15nvidia-nemotron-nano-9b-v2
2.02s
Sep 19, 2026
16Claude Opus 4.6
2.08s
Sep 21, 2026
17Qwen3.5 35B-A3B
2.09s
Sep 21, 2026
18Qwen3.5 397B-A17Bnon-reasoning
2.17s
Sep 20, 2026
19Qwen Turbo
2.18s
Sep 21, 2026
20Qwen3-Next 80B-A3B Instruct
2.20s
Sep 21, 2026
21Qwen3.5 122B-A10Bnon-reasoning
2.34s
Sep 21, 2026
22GLM-5.3-Flash
2.53s
Sep 21, 2026
23GLM-4.5Vthinking
2.82s
Sep 20, 2026
24GLM-4.6V
2.91s
Sep 21, 2026
25Qwen3-Coder 480B-A35B Instruct
3.12s
Sep 21, 2026
26Qwen3.6 27Bnon-reasoning
3.72s
Sep 21, 2026
27Qwen3.5 27Bnon-reasoning
5.60s
Aug 27, 2026
28o3-mini
6.03s
Sep 21, 2026
29o3
6.36s
Sep 21, 2026
30Mercury 2
6.51s
Sep 21, 2026
31GPT-5.1 Codex
8.92s
Jul 25, 2026
32Gemini 3.5 Flash Lite
11.54s
Sep 21, 2026
33Gemini 2.5 Flashthinking
13.98s
Sep 21, 2026
34Muse Spark 1.2
16.10s
Sep 21, 2026
35Claude Opus 5xhigh-effort
22.87s
Sep 21, 2026
36Gemini 2.5 Pro
23.19s
Sep 21, 2026
37GPT-5.1
33.52s
Sep 1, 2026
38Gemini 3.5 Flash
33.65s
Sep 21, 2026
39GPT-5.2 Codex
47.86s
Jul 25, 2026
40GPT-5
62.76s
Sep 21, 2026
41GPT-5.5
71.28s
Sep 21, 2026
42GPT-5 Mini
125.83s
Sep 21, 2026
43GPT-6 Astraxhigh-effort
126.92s
Sep 21, 2026
44Claude Sonnet 5
133.01s
Sep 21, 2026
45GPT-5.6 Luna
136.36s
Sep 21, 2026
46GPT-5.2
170.83s
Sep 21, 2026
47Claude Fable 5.1
275.35s
Sep 21, 2026

Bars on a 0–275.35 seconds scale. Scores link to model pages; every number keeps its source on the model page.