Back to home
Speed
How fast can each AI model generate output? Ranked by tokens per second throughput.
RankModelScore
123
4
5678
9
10
11
121314151617
1819
20212223242526
27282930
C
Celeris 1celeris/celeris-1
1568.2 t/s
I
K2 Horizon 375B A23Bifm/k2-horizon-375b-a23b
1095.6 t/s
Nemotron 3.5 Super VLnvidia/nemotron-3-5-super-vl
418.2 t/s
Gemini 3.5 Flash Litegoogle/gemini-3-5-flash-lite
409.9 t/s
R
Beam 1.0reflection/reflection-beam
359.8 t/s
Nemotron 3.5 Lightningnvidia/nemotron-3-5-lightning
287.5 t/s
NVIDIA Nemotron 3 Nano 30B A3Bnvidia/nvidia-nemotron-3-nano-30b-a3b
265.4 t/s
Nova Microaws/nova-micro
259.6 t/s
gpt oss 120bopenai/gpt-oss-120b-low
241.4 t/s
Claude Haiku 5.5anthropic/claude-haiku-5-5
235.6 t/s
Qwen3.5 Omni Flashalibaba/qwen3-5-omni-flash
234.4 t/s
Nemotron 3 Nano Omni 30B A3B Rnvidia/nemotron-3-nano-omni-30b-a3b
229.9 t/s
T
Inkling Smallthinking-machines/inkling-small
227.8 t/s
Nova 2.0 Liteaws/nova-2-0-lite
224.8 t/s
I
Ling 3.1 Flashinclusionai/ling-3-1-flash
216.0 t/s
NVIDIA Nemotron Nano 12B v2 VLnvidia/nvidia-nemotron-nano-12b-v2-vl
210.7 t/s
Ministral 3 3Bmistral/ministral-3-3b
204.8 t/s
gpt oss 20bopenai/gpt-oss-20b-low
198.8 t/s
Muse Sparkmeta/muse-spark-open
194.3 t/s
S
Step 5stepfun/step-5
183.2 t/s
S
Sarvam 105Bsarvam/sarvam-105b
182.5 t/s
Muse Wondermeta/muse-wonder
180.5 t/s
Mistral Small 4mistral/mistral-small-4-non-reasoning
178.6 t/s
T
Inklingthinking-machines/inkling
177.4 t/s
NVIDIA Nemotron Nano 9B V2nvidia/nvidia-nemotron-nano-9b-v2-reasoning
175.8 t/s
DeepSeek V4 Flash Visiondeepseek/deepseek-v4-flash-vision
167.2 t/s
DeepSeek V4.1 Flash 1017deepseek/deepseek-v4-1-flash-1017
164.3 t/s
Qwen3.6 35B A3Balibaba/qwen3-6-35b-a3b-non-reasoning
160.9 t/s
M
Quasar 438Bmultiversecomputing/quasar-438b
160.7 t/s
Nemotron 3 Ultra 550B A55Bnvidia/nvidia-nemotron-3-ultra-550b-a55b
159.2 t/s
470 models tested