TPSTokens Per Second

AI Model Benchmarks

500 scores across 12 benchmarks — quality, coding, reasoning, and speed.

Output throughput in tokens per second measured on standard prompts. Higher = faster responses. Source: artificialanalysis.ai, ~Feb 2026.

Model
tok/s
1
Nemotron-Labs Diffusion 8BNVIDIA
865
2
DiffusionGemma 26B A4B ITGoogle DeepMind
500
3
Gemini 3.1 Flash LiteGoogle DeepMind
360
4
Step-3.5-Flash-BaseStepFun
300
5
Gemini 2.0 FlashGoogle DeepMind
228
6
LFM2.5-230MLiquid Ai
213
7
GPT-5.4 NanoOpenAI
210
8
Gemini 2.5 FlashGoogle DeepMind
198
9
Claude Haiku 4.5Anthropic
187
10
SonarPerplexity AI
175
11
Gemma 4 26B A4B ITGoogle DeepMind
163.09
12
Llama 4 ScoutMeta AI
160
13
Mistral SabaMistral AI
150
14
Grok 4 minixAI
145
15
Grok 3 minixAI
138
16
GPT-5.4 miniOpenAI
132
17
Sonar ReasoningPerplexity AI
120
18
o4-miniOpenAI
115
19
GPT-4.1OpenAI
112
20
o3-miniOpenAI
108
21
Ornith-1.0Deepreinforce
103
22
GPT-4oOpenAI
98
23
Sonar ProPerplexity AI
95
24
Qwen3.5-35B-A3BAlibaba / Qwen
92.51
25
Llama 4 MaverickMeta AI
90
26
GPT-5.4OpenAI
85
27
Claude Sonnet 4.5Anthropic
82
28
Claude Sonnet 4.6Anthropic
80
29
Claude 3.5 SonnetAnthropic
78
30
Mistral Medium 3Mistral AI
75
31
Mistral Large 2Mistral AI
72
32
DeepSeek V3DeepSeek
70
33
Claude 3.7 SonnetAnthropic
68
34
Sonar Reasoning ProPerplexity AI
65
35
Qwen3 72BAlibaba / Qwen
65
36
Gemini 2.5 ProGoogle DeepMind
62
37
Phi-4Microsoft
60
38
DeepSeek R1DeepSeek
58
39
Grok 3xAI
55
40
QwQ-32BAlibaba / Qwen
50
41
Gemini 1.5 ProGoogle DeepMind
45
42
GPT-4.5OpenAI
42
43
Claude Opus 4Anthropic
40
44
Claude Opus 4.6Anthropic
38
45
o3OpenAI
32
46
Grok 4xAI
28
47
Llama 3.1 405BMeta AI
26
48
Grok 4.20xAI
24
49
o1OpenAI
22
50
Sonar Deep ResearchPerplexity AI
18

Quality scores from official model cards and published technical reports. Speed benchmarks from artificialanalysis.ai (approximate medians). Higher is better for all metrics.