Benchmarks / Artificial Analysis

Reported by Artificial Analysis

Artificial Analysis

Median tokens a second while the model writes its answer, measured by Artificial Analysis on the model's usual API over the past weeks.

Last updated 20 Oct 2026

Results dated
20 Oct 2026
Results
137 configurations of 78 models
Unit
tokens per second
Licence
Artificial Analysis commercial data licence

Output speed: Inkling

Top 15 of 78 results · tokens per second, higher is better. Choose a model to highlight it.Clear highlight

  1. 1 Nemotron 3.5 Super VLNVIDIA 411
  2. 2 Gemini 3.5 Flash-LiteGoogle 398
  3. 3 DeepSeek-V4.1-Flash (no reasoning)DeepSeek 369
  4. 4 Trinity Large ThinkingArcee AI 362
  5. 5 Beam 1.0 501B A23BReflection 323
  6. 6 gpt-oss-20b (low reasoning)OpenAI 277
  7. 7 Nova Micro 1.0Amazon 260
  8. 8 gpt-oss-120b (high reasoning)OpenAI 259
  9. 9 Muse Spark 1.3 (extra-high reasoning)Meta 254
  10. 10 Claude Haiku 5.5 (max reasoning)Anthropic 240
  11. 11 Nemotron 3 Nano 30B A3B (reasoning on)NVIDIA 238
  12. 12 Nova 2 Lite (no reasoning)Amazon 233
  13. 13 Inkling (extra-high reasoning)Thinking Machines 232
  14. 14 Qwen3.5-Omni-FlashAlibaba 231
  15. 15 DeepSeek-V4.1-Flash-1017 (max reasoning)DeepSeek 212

Full results

Artificial Analysis: Output speed, tokens per second, higher is better
#ModelOutput speed
tokens per second, higher is better		Price
$ per million tokens, in / out
1 Nemotron 3.5 Super VLNVIDIA
411
–
2 Gemini 3.5 Flash-LiteGoogle
398
$0.30 / $2.50
3 DeepSeek-V4.1-Flash (no reasoning)DeepSeek · best of 2 settings
369
$0.30 / $1.20
4 Trinity Large ThinkingArcee AI
362
$0.25 / $0.80
5 Beam 1.0 501B A23BReflection
323
–
6 gpt-oss-20b (low reasoning)OpenAI · best of 2 settings
277
$0.03 / $0.15
7 Nova Micro 1.0Amazon
260
$0.035 / $0.14
8 gpt-oss-120b (high reasoning)OpenAI · best of 2 settings
259
$0.15 / $0.60
9 Muse Spark 1.3 (extra-high reasoning)Meta · best of 2 settings
254
$1.25 / $4.25
10 Claude Haiku 5.5 (max reasoning)Anthropic · best of 5 settings
240
$0.10 / $0.50
11 Nemotron 3 Nano 30B A3B (reasoning on)NVIDIA · best of 2 settings
238
$0.05 / $0.20
12 Nova 2 Lite (no reasoning)Amazon · best of 4 settings
233
$0.30 / $2.50
13 Inkling (extra-high reasoning)Thinking Machines
232
$0.95 / $4.05
14 Qwen3.5-Omni-FlashAlibaba
231
–
15 DeepSeek-V4.1-Flash-1017 (max reasoning)DeepSeek
212
$0.30 / $1.20
16 Ministral 3 3BMistral AI
209
$0.10 / $0.10
17 DeepSeek-V4-Flash-Vision-Exp (max reasoning)DeepSeek
191
$0.44 / $1.32
18 Inkling SmallThinking Machines
181
$0.45 / $1.20
19 Mistral Small 4Mistral AI · best of 2 settings
176
$0.15 / $0.60
20 Qwen3-Next-80B-A3B-ThinkingAlibaba
163
$0.15 / $1.20
21 MiniMax M3.1 FlashMiniMax
159
$0.15 / $0.60
22 Mistral Medium 3.5Mistral AI
157
$1.50 / $7.50
23 o3OpenAI
155
$2 / $8
24 Muse Wonder (max reasoning)Meta
153
$4 / $20
25 GPT-5.5 Instant (2026-06-26)OpenAI
152
–
26 Gemma 4 12BGoogle · best of 2 settings
148
–
27 MiniMax-M3MiniMax
146
$0.30 / $1.20
28 Qwen3.5-122B-A10B (no reasoning)Alibaba · best of 2 settings
144
$0.26 / $2.08
29 Muse Glimmer (high reasoning)Meta
142
–
29 Muse Spark Open (extra-high reasoning)Meta
142
–
31 Gemini 3.8 Flash (high reasoning)Google
141
$1.50 / $7.50
32 GPT-5.6 Terra (max reasoning)OpenAI · best of 6 settings
140
$2 / $12
33 Claude Sonnet 5.5 (max reasoning)Anthropic · best of 5 settings
137
$2 / $10
34 Qwen3.6-35B-A3B (no reasoning)Alibaba · best of 2 settings
135
$0.10 / $1
35 Llama 4 ScoutMeta
132
$0.18 / $0.59
36 Gemini 3.1 Pro PreviewGoogle
130
$2 / $12
37 GPT-6 Luna (max reasoning)OpenAI · best of 5 settings
127
$0.10 / $0.50
38 Qwen3-Next-80B-A3B-InstructAlibaba
121
$0.10 / $1.10
39 GPT-5.3-Codex (extra-high reasoning)OpenAI
119
$1.75 / $14
39 Solar Pro 4Upstage
119
$0.09 / $0.36
41 Qwen3.5-9B (no reasoning)Alibaba · best of 2 settings
110
$0.10 / $0.15
42 Qwen3-Omni-30B-A3B-ThinkingAlibaba
108
–
43 Qwen3-Omni-30B-A3B-InstructAlibaba
107
–
44 Llama 3.3 70B InstructMeta
97.7
$0.59 / $0.79
45 Qwen3.5-Omni-PlusAlibaba
96.5
–
46 Claude Opus 5.5 (max reasoning)Anthropic · best of 5 settings
96.0
$4 / $20
47 DeepSeek-V4-Pro (0813, max reasoning)DeepSeek · best of 2 settings
95.9
$1.32 / $3.96
48 Ministral 3 8BMistral AI
93.2
$0.15 / $0.15
49 GLM-5.3 (low reasoning)Z.ai · best of 2 settings
87.5
$1.40 / $4.40
50 Hy3Tencent
83.3
$0.14 / $0.58
50 Reka Flash 3Reka AI
83.3
$0.10 / $0.20
52 Gemma 4 26B A4B (no reasoning)Google · best of 2 settings
81.8
$0.10 / $0.30
53 Ministral 3 14BMistral AI
79.2
$0.20 / $0.20
54 GLM-5.3-FlashZ.ai
78.7
$0.15 / $0.50
55 Qwen3-Coder-NextAlibaba
77.3
$0.18 / $0.90
56 Magistral Small 1.2Mistral AI
76.0
–
57 Grok 4.7 (extra-high reasoning)xAI · best of 3 settings
74.3
$2 / $6
58 Claude Fable 5.1 (max reasoning)Anthropic · best of 5 settings
72.5
$10 / $50
59 GPT-6 Astra (max reasoning)OpenAI · best of 5 settings
71.7
$10 / $50
60 Granite 4.2 8BIBM
64.9
$0.06 / $0.25
61 Mistral Large 4Mistral AI
60.2
$1.36 / $4.18
62 Command ACohere
59.5
$2.50 / $10
63 Qwen3.7-PlusAlibaba
56.1
$0.32 / $1.28
64 Qwen3.5-397B-A17B (no reasoning)Alibaba · best of 2 settings
55.6
$0.55 / $3.50
65 Qwen3.8-Flash-NextAlibaba
55.0
–
66 Kimi K2.7 CodeMoonshot AI
53.7
$0.95 / $4
67 GPT-6.1 Sol (max reasoning)OpenAI · best of 5 settings
53.1
$2 / $10
68 Gemma 4 31B (no reasoning)Google · best of 2 settings
47.6
$0.14 / $0.40
69 Qwen3.8-27B (low reasoning)Alibaba · best of 4 settings
46.4
$0.50 / $3
69 Llama 4 MaverickMeta
46.4
$0.27 / $0.85
71 MiMo-V2.6-FlashXiaomi
46.2
$0.14 / $0.28
72 Phi-4Microsoft
45.1
$0.07 / $0.14
73 MiMo-V2.6-ProXiaomi
42.7
$0.43 / $0.87
74 Kimi K3 (max reasoning)Moonshot AI · best of 2 settings
41.0
$3 / $15
75 Qwen3.8-Max (0902)Alibaba
36.8
$2 / $6
76 Qwen3.8-2.4T-A95BAlibaba
35.7
$2 / $6
77 Gemma 4 E4B (no reasoning)Google · best of 2 settings
31.7
–
78 Qwen3.5-4B (reasoning on)Alibaba · best of 2 settings
9.71
–

Swipe the table sideways for more columns.

Ranks follow the score as shown, so equal numbers share a rank. Each model is shown at its best setting; show every setting. Results as published by Artificial Analysis; we do not re-run them.

Progress over time

Output speed, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

0200400 2025Jul2026Jul Nova Micro 1.0gpt-oss-20bTrinity Large ThinkingGemini 3.5 Flash-LiteNemotron 3.5 Super VL

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 17 others not drawn.

NVIDIAGoogleDeepSeekArcee AIReflectionOpenAI

0100200300400500 2025Jul2026Jul NVIDIAGoogleDeepSeekArcee AIReflectionOpenAI

What it measures

Median tokens a second while the model writes its answer, measured by Artificial Analysis on the model's usual API over the past weeks.

What it does not measure

Not time spent thinking before the answer starts, or speed on other hosts.

Source

Artificial Analysis · Methodology

285 results from Artificial Analysis not ranked here · show why

We rank a result only when we can tie it to a specific model you can use. These are left out:

  • Not on sale through the API providers we track: 270
  • A different snapshot or variant from the model we list: 14
  • An unusual combination of settings: 1

Data sourced from Artificial Analysis. Licence: Artificial Analysis commercial data licence.