AI Parameter Counts

A parameter count is the number of weights in an AI model. Total parameters are every weight in the network. Active parameters are the weights used for one token in a mixture-of-experts model. Dense models use every weight, so the two counts match. The size score places a published total on a 100-point log scale from 1 billion to 10 trillion.

Data sourced from OpenRouter API and manual data collection | Last updated: Nov 3, 2026, 8:01 AM UTC

Mixture-of-experts models

Model Name	Total	Active	Size Score
Qwen: Qwen3.8 2.4T A95B	2.4T	95B	85
Meituan: LongCat 2.0	1.6T	48B	80
Mistral: Mistral Large 4	1T	52B	75
MoonshotAI: Kimi K2 0711	1T	32B	75
Thinking Machines: Inkling	975B	41B	75
Meta: Muse Spark Open	816B	42B	73
Tencent: Hy4 preview	770B	49B	72
Mistral: Mistral Large 3 2512	675B	41B	71
DeepSeek: DeepSeek V3.1	671B	37B	71
DeepSeek: R1 0528	671B	37B	71
StepFun: Step 5 Preview	600B	27B	69
inclusionAI: Ling 3.1 Flash	560B	25B	69
NVIDIA: Nemotron 3 Ultra	550B	55B	69
Reflection: Beam 1.0 501B A23B	501B	23B	67
Qwen: Qwen3 Coder 480B A35B	480B	35B	67
Qwen: Qwen3.5 397B A17B	397B	17B	65
Tencent: Hy3	295B	21B	62
DeepSeek: DeepSeek V4 Flash 0731	284B	13B	61
Thinking Machines: Inkling Small	276B	12B	61
Qwen: Qwen3 235B A22B Instruct 2507	235B	22B	59
Qwen: Qwen3 VL 235B A22B Instruct	235B	22B	59
Qwen: Qwen3 235B A22B Thinking 2507	235B	22B	59
Mistral: Mixtral 8x22B Instruct	176B	39B	56
inclusionAI: Ling 3.0 Flash VL	124B	5.5B	52
Qwen: Qwen3.5-122B-A10B	122B	10B	52
NVIDIA: Nemotron 3.5 Super VL	120B	12B	52
NVIDIA: Nemotron 3 Super	120B	12B	52
Qwen: Qwen3 Next 80B A3B Instruct	80B	3B	48
Upstage: Solar Mini 4	35B	3B	39
Qwen: Qwen3.6 35B A3B	35B	3B	39
Qwen: Qwen3.5-35B-A3B	35B	3B	39
Qwen: Qwen3 30B A3B Instruct 2507	30.5B	3.3B	37
Qwen: Qwen3 Coder 30B A3B Instruct	30.5B	3B	37
NVIDIA: Nemotron 3.5 Lightning	30B	3B	37
Qwen: Qwen3 VL 30B A3B Instruct	30B	3B	37
NVIDIA: Nemotron 3 Nano 30B A3B	30B	3B	37
Qwen: Qwen3 VL 30B A3B Thinking	30B	3B	37
Qwen: Qwen3 30B A3B	30B	3B	37
Google: Gemma 4 26B A4B	25.2B	3.8B	35
OpenAI: gpt-oss-20b	21B	3.6B	33
Tencent: Hunyuan A13B Instruct	Undisclosed	13B	28
Moondream: Moondream 3 Preview	9B	2B	24
Moondream: Moondream 3.1	9B	2B	24
LiquidAI: LFM2.5-24B-A2B	Undisclosed	2.45B	10

🤖 Generate website with AI 👆

Join the beta waitlist and get 50% off your first 3 months.

Join waitlist

Dense models

Model Name	Total	Active	Size Score
MoonshotAI: Kimi K3	2.8T	Undisclosed	86
DeepSeek: DeepSeek V4 Pro 0813	1.6T	Undisclosed	80
Xiaomi: MiMo-V2.6-Pro	1T	Undisclosed	75
MoonshotAI: Kimi K2 0905	1T	Undisclosed	75
DeepSeek: DeepSeek V3 0324	685B	Undisclosed	71
MiniMax: MiniMax-01	456B	Undisclosed	66
Nous: Hermes 4 405B	405B	Undisclosed	65
Xiaomi: MiMo-V2.6-Flash	309B	Undisclosed	62
MiniMax: MiniMax M2	230B	Undisclosed	59
StepFun: Step 3.7 Flash	196B	Undisclosed	57
StepFun: Step 3.5 Flash	196B	Undisclosed	57
inclusionAI: Ling 3.0 Flash	124B	Undisclosed	52
Mistral: Devstral 2 2512	123B	Undisclosed	52
Mistral: Mistral Small 4	119B	Undisclosed	52
OpenAI: gpt-oss-120b	117B	Undisclosed	52
Cohere: Command A	111B	Undisclosed	51
Meta: Llama 4 Scout	109B	Undisclosed	51
Z.ai: GLM 4.5V	106B	Undisclosed	51
Qwen: Qwen3 Coder Next	80B	Undisclosed	48
Qwen: Qwen2.5 VL 72B Instruct	72B	Undisclosed	46
Qwen2.5 72B Instruct	72B	Undisclosed	46
Magnum v4 72B	72B	Undisclosed	46
Meta: Llama 3.3 70B Instruct	70B	Undisclosed	46
Meta: Llama 3.1 70B Instruct	70B	Undisclosed	46
Sao10K: Llama 3.1 Euryale 70B v2.2	70B	Undisclosed	46
TheDrummer: Skyfall 36B V2	36B	Undisclosed	39
Qwen: Qwen3 32B	32.8B	Undisclosed	38
Qwen2.5 Coder 32B Instruct	32B	Undisclosed	38
Google: Gemma 4 31B	31B	Undisclosed	37
Meta: Muse Glimmer 30B	30B	Undisclosed	37
PrismML: Ternary Bonsai 2 27B	27B	Undisclosed	36
Qwen: Qwen3.8 27B	27B	Undisclosed	36
Qwen: Qwen3.6 27B	27B	Undisclosed	36
Qwen: Qwen3.5-27B	27B	Undisclosed	36
Google: Gemma 3 27B	27B	Undisclosed	36
Google: Gemma 2 27B	27B	Undisclosed	36
Mistral: Mistral Small 3.2 24B	24B	Undisclosed	35
TheDrummer: Cydonia 24B V4.1	24B	Undisclosed	35
Venice: Uncensored	24B	Undisclosed	35
Mistral: Mistral Small 3	24B	Undisclosed	35
Mistral: Saba	24B	Undisclosed	35
Mistral: Mistral Small 3.1 24B	24B	Undisclosed	35
Mistral: Voxtral Small 24B 2507	24B	Undisclosed	35
Reka Flash 3	21B	Undisclosed	33
OpenAI: gpt-oss-safeguard-20b	21B	Undisclosed	33
Meta: Llama 4 Maverick	17B	17B	31
Qwen: Qwen3 14B	14.8B	Undisclosed	29
Mistral: Ministral 3 14B 2512	14B	Undisclosed	29
MythoMax 13B	13B	Undisclosed	28
ReMM SLERP 13B	13B	Undisclosed	28
Google: Gemma 3 12B	12B	Undisclosed	27
TheDrummer: UnslopNemo 12B	12B	Undisclosed	27
Mistral: Mistral Nemo	12B	Undisclosed	27
Meta: Llama Guard 4 12B	12B	Undisclosed	27
PrismML: Ternary Bonsai 2 9B	9B	Undisclosed	24
Qwen: Qwen3.5-9B	9B	Undisclosed	24
DeepSeek: DeepSeek V4.1 Flash 1017	8B	Undisclosed	23
IBM: Granite 4.2 8B	8B	Undisclosed	23
Meta: Llama 3.1 8B Instruct	8B	Undisclosed	23
Mistral: Ministral 3 8B 2512	8B	Undisclosed	23
Qwen: Qwen3 VL 8B Instruct	8B	Undisclosed	23
Sao10K: Llama 3 8B Lunaris	8B	Undisclosed	23
AionLabs: Aion-RP 1.0 (8B)	8B	Undisclosed	23
ByteDance: UI-TARS 7B	7B	Undisclosed	21
Qwen: Qwen2.5 7B Instruct	7B	Undisclosed	21
Google: Gemma 3 4B	4B	Undisclosed	15
Meta: Llama 3.2 3B Instruct	3B	Undisclosed	12
Mistral: Ministral 3 3B 2512	3B	Undisclosed	12
LiquidAI: LFM2.5-2.6B	2.6B	Undisclosed	10
Meta: Llama 3.2 1B Instruct	1B	Undisclosed	0

AI Parameter Counts

Mixture-of-experts models

🤖 Generate website with AI 👆

Dense models

🤖 Generate website with AI 👆

Frequently asked questions

What is an AI model parameter count?