#ai tech

- Llama 3.1 — https://ollama.com/library/llama3.1

- Llama 3.2 — https://ollama.com/library/llama3.2

- Nomic Embed Text — https://ollama.com/library/nomic-embed-text

- GPT-OSS — https://ollama.com/library/gpt-oss

- Qwen 3.5 — https://ollama.com/library/qwen3.5

- CodeQwen — https://ollama.com/library/codeqwen

- OpenHermes — https://ollama.com/library/openhermes

- MedLlama2 — https://ollama.com/library/medllama2

- EverythingLM — https://ollama.com/library/everythinglm

- Olmo — https://ollama.com/library/olmo

General Purpose / Chat Models

Llama 3.1 — https://ollama.com/library/llama3.1

Meta’s flagship 8B, 70B, 405B models. Strong reasoning and instruction following. Best all-rounder.

Llama 3.2 — https://ollama.com/library/llama3.2

1B and 3B variants optimized for edge/mobile. Also 11B and 90B vision models for image input.

Llama 3.3 — https://ollama.com/library/llama3.3

Newest 70B only. Better than Llama 3.1 70B at math, coding, and following long prompts.

Mistral — https://ollama.com/library/mistral

7B v0.3. Fast, good for English tasks. Great baseline for 6-8GB VRAM.

Mixtral — https://ollama.com/library/mixtral

8x7B and 8x22B MoE models. 8x7B runs like a 12B but performs near 70B models.

Phi-3.5 — https://ollama.com/library/phi3.5

3.8B Mini, 14B. Microsoft models. Excellent at JSON, code, and reasoning for their size.

Gemma 2 — https://ollama.com/library/gemma2

Google 2B, 9B, 27B. 9B punches above its weight. Strong safety tuning.

Qwen 2.5 — https://ollama.com/library/qwen2.5

0.5B to 72B range. 7B and 14B are best for 6GB and 16GB VRAM. Excellent multilingual + JSON.

Qwen 3 — https://ollama.com/library/qwen3

Newest Qwen family. 0.6B to 235B. 8B and 14B versions are very strong for reasoning.

Coding Models

CodeLlama — https://ollama.com/library/codellama

7B, 13B, 34B, 70B. Meta code specialist. Fill-in-the-middle support.

CodeQwen — https://ollama.com/library/codeqwen

7B code model based on Qwen2. Stronger than CodeLlama 7B on benchmarks.

DeepSeek Coder V2 — https://ollama.com/library/deepseek-coder-v2

16B and 236B MoE. Best open code model right now. 16B runs on 24GB VRAM.

Codestral — https://ollama.com/library/codestral

22B by Mistral. Trained on 80+ languages. Fast for autocomplete.

Reasoning / Instruction Tuned

OpenHermes 2.5 — https://ollama.com/library/openhermes

7B Mistral finetune. Very good at following complex instructions.

Nous Hermes 2 — https://ollama.com/library/nous-hermes2

7B, 34B variants. Uncensored and strong at roleplay + reasoning.

Dolphin Llama 3 — https://ollama.com/library/dolphin-llama3

8B and 70B. Uncensored finetunes of Llama 3. Good for creative tasks.

EverythingLM — https://ollama.com/library/everythinglm

13B Llama 2 finetune. Trained to be helpful and detailed.

Embedding / Retrieval Models

Nomic Embed Text — https://ollama.com/library/nomic-embed-text

137M. Best open embedding model for RAG. 8192 context.

mxbai-embed-large — https://ollama.com/library/mxbai-embed-large

335M. Strong MTEB score. Alternative to Nomic.

bge-m3 — https://ollama.com/library/bge-m3

567M. Multi-lingual, multi-function, multi-granularity embeddings.

Domain Specific

MedLlama2 — https://ollama.com/library/medllama2

7B medical finetune of Llama 2. For clinical QA, not diagnosis.

Meditron — https://ollama.com/library/meditron

7B and 70B medical LLMs. Trained on PubMed + guidelines.

BioMistral — https://ollama.com/library/biomistral

7B Mistral finetune for biomedical domain.

Science / Research

Olmo — https://ollama.com/library/olmo

7B by AI2. Fully open data, code, weights. Good for research.

Yi — https://ollama.com/library/yi

6B, 9B, 34B. Bilingual Chinese-English. Strong reasoning.

Multimodal / Vision

LLaVA — https://ollama.com/library/llava

7B, 13B, 34B. Llama + CLIP. Chat about images.

BakLLaVA — https://ollama.com/library/bakllava

Mistral 7B + LLaVA. Better vision than base LLaVA.

Moondream 2 — https://ollama.com/library/moondream

1.9B tiny vision model. Runs on CPU. Good for captions.

New / Experimental

GPT-OSS — https://ollama.com/library/gpt-oss

Open weights from OpenAI, 20B MoE. Apache 2.0 license.

Command R — https://ollama.com/library/command-r

35B by Cohere. Optimized for RAG and tool use.

Command R+ — https://ollama.com/library/command-r-plus

104B by Cohere. Best open model for agents and long context.

How to pull any of these:

`ollama pull llama3.1:8b-instruct-q4_K_M`

Replace the name and tag with what you need. Add `:q4_K_M` for best VRAM/quality balance on 6-24GB GPUs.

For your RTX 4050 6GB: Stick to 7B-8B Q4_K_M or smaller: llama3.1:8b, qwen2.5:7b, phi3.5:3.8b, gemma2:9b-instruct-q3_K_M.

RUBEM DIDINI FILHO

We don't just build software—we build solutions that drive growth.

✨ AI Development 🔗 Blockchain Solutions 🌐 Web3 Applications 💻 Custom Software

🚀 Delivered on time. Every time.

💡 Quick Take: SM Entertainment stands at the forefront of the digital revolution, leveraging cutting-edge AI, a key component of SM Entertainment AI explained, to create hyper-realistic K-Pop virtual idols and immersive fan experiences that are redefining entertainment. 1. The Big Picture: Why This Matters Now 1.1. Global Context The global virtual influencer market, valued at approximately $4.6…

1. Global Trigger: The Macro Shift The global technology landscape is cementing itself around two distinct and competing spheres of influence: the US-led alliance and the self-reliant push from Beijing. We are seeing clear evidence of this strategic bifurcation in trade flows and policy pronouncements. Recent data shows US exports from Taiwan now exceeding those from China for the first time in…

-- Yann LeCun, "godfather of AI", in response to a comment under his Facebook post about the delusional e-mails he gets from all kinds of fuckwits.