BenchLeader

Latency: time to first answer

Median seconds until the answer starts, including any reasoning phase (Artificial Analysis), or time to first streamed token where that is all we have.

As of 15 Nov 2026, BenchLeader ranks Tiny Aya Global first on the Latency: time to first answer board at 0.20 s, ahead of Llama 3.2 3B Instruct (0.21 s) and Llama 3.2 1B Instruct (0.22 s).

706 models · data as of 15 Nov 2026

Top 15 by seconds
  1. 1Tiny Aya Global0.20 s
  2. 2Llama 3.2 3B Instruct0.21 s
  3. 3Llama 3.2 1B Instruct0.22 s
  4. 4Nemotron 3 Ultra 550B A55B0.25 s
  5. 5Hermes 4 405B (none)0.30 s
  6. 6Beam 1.0 501B A23B0.32 s
  7. 7Codestral 25080.32 s
  8. 8GLM 4.7 Flash0.33 s
  9. 9gpt-oss-20b0.33 s
  10. 10Nemotron 3 Nano 30B A3B0.34 s
  11. 11gpt-oss-120b0.35 s
  12. 12Gemini 2.5 Flash-Lite (none)0.36 s
  13. 13Inkling0.39 s
  14. 14Phi 4 Multimodal0.39 s
  15. 15Qwen3 32B0.40 s
262 of 262
#
1Kimi K3.1 Pro178.90 s30 tok/s
2Qwen3.8 2.4T A95B119.16 s39 tok/s
3Hy4115.63 s92 tok/s
4GLM 5.3 Flash102.11 s50 tok/s
5Solar Mini 490.67 s96 tok/s
6Qwen3.8-Flash-Next87.24 s77 tok/s
7MiniMax M379.04 s118 tok/s
8Inkling 2 Small68.69 s88 tok/s
9DeepSeek R1 052864.09 s30 tok/s
10Granite 4.2 8B62.49 s66 tok/s
11Qwen3.7 Plus54.40 s58 tok/s
12LongCat 2.053.88 s61 tok/s
13MiniMax M3.1 Flash50.19 s147 tok/s
14Ling 3.1 Flash50.01 s90 tok/s
15Qwen3.6 Plus48.82 s56 tok/s
16Hy347.80 s76 tok/s
17Deepseek v4 Flash Visionmax47.69 s159 tok/s
18Seed-OSS-36B-Instruct45.40 s34 tok/s
19Kimi K2.7 Code45.31 s54 tok/s
20MiMo-V2.6-Pro43.57 s56 tok/s
21MiMo-V2.6-Flash39.29 s46 tok/s
22Solar Pro 431.39 s78 tok/s
23MiniMax M2.528.28 s40 tok/s
24Step 3.5 Flash 260327.04 s127 tok/s
25Mistral Large 4medium25.82 s86 tok/s
26Step 3.5 Flash 020225.28 s131 tok/s
27Mistral Medium 3.523.91 s147 tok/s
28Step 5 Preview23.76 s128 tok/s
29Kimi K2 Thinkingthinking20.90 s127 tok/s
30o4-minihigh19.10 s116 tok/s
31Apodex 1.118.94 s112 tok/s
32Command A+18.30 s183 tok/s
33MiniMax M2.717.90 s43 tok/s
34K2 Horizon 375B A23B17.71 s244 tok/s
35Gemini 2.5 Pro17.49 s110 tok/s
36Nemotron 3.5 Super VL17.14 s226 tok/s
37Grok 4.20 Multi-Agent16.72 s316 tok/s
38Grok 416.70 s67 tok/s
39Gemini 3.1 Prohigh15.56 s112 tok/s
40Inkling Small15.14 s117 tok/s
41Step 3.7 Flash15.05 s139 tok/s
42Qwen3.7max15.03 s170 tok/s
43GLM-4.5-Air14.35 s68 tok/s
44GPT-5.2 Proxhigh13.92 s9 tok/s
45DeepSeek R113.84 s105 tok/s
46Nemotron 3.5 Lightning13.34 s269 tok/s
47Ling 3.0 Flash12.96 s116 tok/s
48Solar Pro 312.58 s126 tok/s
49Ling 3.0 Tiny11.97 s180 tok/s
50Grok Build 0.111.38 s169 tok/s
51Sarvam 105Bhigh11.06 s91 tok/s
52Ring-2.6-1T10.34 s107 tok/s
53DeepSeek V4 Flash 0420max9.67 s93 tok/s
54GPT-5-Codexhigh9.46 s123 tok/s
55GPT-5.4 Pro8.80 s6 tok/s
56Llama 3.1 70B8.68 s17 tok/s
57Command A8.59 s44 tok/s
58Qwen3 Coder 30B A3B Instruct8.20 s180 tok/s
59MiniMax-M28.09 s97 tok/s
60Mercury 26.98 s349 tok/s

Cite as: BenchLeader, “Latency: time to first answer”, data as of 15 Nov 2026, https://www.benchleader.com/leaderboards/latency.