ResearchLLM Verifications Leaderboard

Data as of · 78 models, 80 questions · reviewed by Suhas Bhairav · methodology

Best LLM for…

The LLMs with the lowest latency

Answer: Ministral 3 3B 2512 by Mistral, with a median 0.3 s to the first token.

Reviewed by Suhas Bhairav · Data as of

Ranking

Median time from sending the request to the first streamed token of any kind, so thinking time counts. Top 15 of 74.

The LLMs with the lowest latency
#ModelFirst tokenSpeed (tok/s)ScoreCost / question
1Ministral 3 3B 2512Mistral0.3 s25226.2%$0.00019
2Ministral 3 8B 2512Mistral0.3 s12442.5%$0.00027
3Hermes 4 405BNous0.3 s4126.2%$0.01
4Kimi K2.7 CodeMoonshotAI0.4 s13665.0%$0.006
5MiMo-V2.6-FlashXiaomi0.4 s7762.5%$0.00056
6Qwen3.6 35B A3BQwen0.4 s13537.5%$0.0037
7Qwen3 Coder NextQwen0.4 s10662.5%$0.0017
8Mistral Medium 3.5Mistral0.5 s14528.8%$0.025
9Phi 4Microsoft0.5 s10341.2%$0.00012
10Nova Micro 1.0Amazon0.5 s18022.5%$0.000087
11Reka Flash 3rekaai0.5 s1011.2%$0.0008
12Mistral Small 4Mistral0.5 s17032.5%$0.0019
13Nemotron 3.5 LightningNVIDIA0.5 s11643.8%$0.00049
14Muse Spark OpenMeta0.5 s22077.5%$0.0045
15GLM 5.3 FlashZ.ai0.5 s8367.5%$0.00015

How to read this

Scores come from 80 multiple-choice questions from SuperGPQA and MMLU-Pro. With that many questions, a score has a margin of error of several points, so models a few points apart are effectively tied on accuracy; the comparison pages run a significance test for any pair. To see the top two side by side, compare Ministral 3 3B 2512 and Ministral 3 8B 2512.

Speed, latency and cost are measured through OpenRouter and depend on which host serves the request; your numbers on a direct API may differ. The full procedure is on the methodology page, and the data behind this ranking can be downloaded.

Other questions

Quick answers

Which AI model starts responding the quickest?
Ministral 3 3B 2512 by Mistral, with a median 0.3 s to the first token. Rule: Median time from sending the request to the first streamed token of any kind, so thinking time counts.
What is the runner-up?
Ministral 3 8B 2512 by Mistral, with a median 0.3 s to the first token.
Which of these costs least per correct answer?
gpt-oss-20b: $0.00013 per correct answer (cost per question divided by its score).