0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

LLMのContext windowサイズと画像対応

Last updated at Posted at 2025-06-05
context window 画像対応 4ビット量子化モデルサイズ GPQA AIME 2025
cogito 70b 128K x 43GB
Command-a 16k x 67GB
devstral-small-2 384K 15GB
Gemma3 4B 128k 3.3GB
Gemma3 12B 128k 8.1GB
Gemma3 27B 128k 17GB
Gemma3n:e4b 32k x 7.5GB
gpt-oss:20b 128k x 14GB 74.2%(reasoning_effort=high), 71.5% 98.7%(high)
gpt-oss:120b 128k x 65GB 80.9%(high), 80.1% 92.5%(high)
granite4:micro-h 1M x 1.9GB
granite4:tiny-h 1M x 4.2GB
granite4:small-h 1M x 19GB
Llama3.3 70B 128k x 128K
Llama4 16x17b 10M 67GB
mistral-small3.2 128k 15GB
mistral-large 128k x 73GB
mistral3:14b 256k 9.1GB 71.2% 85.0%
nemotron-3-nano:30b 1M x 24GB 75.0%
Phi4-reasoning 32k x 11GB
Qwen2.5 VL 72B 125k 49GB
Qwen3 4B 256k x 2.5GB
Qwen3 8B 40k x 5.2GB
Qwen3 32B 40k x 20GB
Qwen3 30B-a3b 40k x 19GB
Qwen3 30B-a3b thinking 256k x 19GB 73.4% 85.0%
Qwen3-coder:30b 256k x 19GB
qwen3-next:80b 256K x 50GB 72.9%
qwen3-next:80b-a3b-thinking 256K x 50GB 77.2% 87.8%
qwen3-vl:8b-thinking 256K 6.1GB 69.9% 80.3%
qwen3-vl:30b-a3b-thinking 256K 20GB 74.4% 83.1%
qwen3-vl:32b-instruct 256K 21GB 68.9%
qwen3-vl:32b-thinking 256K 21GB 73.1% 83.7%
rnj-1 32K x 5.1GB
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?