Hello.
Right at the start of April, we are announcing new versions of AI models.
Gemma 4
You've probably already heard about Google Gemma 4.
Gemma 4 model overview (https://ai.google.dev/gemma/docs/core?hl=ko)
With 32GB RAM + RTX 4060 Ti (8GB), you can use up to 26B with Ollama (CPU offloading). The response speed seems a bit slower than nemotrao-cascade-2, probably because Thinking is longer.
31B model failed due to insufficient memory!!
Qwen 3.6 Plus (Free)
The plus model supports vision. It supports 1 million tokens and the API is available for free.
It received the "Best Value" award on the AI Model Leader Rank site (because it's free?).
https://lmmarketcap.com/model/qwen-qwen3-6-plus-free
GLM-5V-Turbo
zAI's first 'multimodal coding model' - is this used for coding by taking images, videos, and files as input? The benchmark results say it performs better than Opus 4.6. Since it's paid, it will be difficult to try it out.
https://docs.z.ai/guides/vlm/glm-5v-turbo#multimodal-coding-and-agentic-tasks
