Machine Learning Models - spinningideas/resources GitHub Wiki

A machine learning model is a file that contains the result of training that utilizes a defined algorithm to recognize certain types of patterns in the data set that is used to train the model. The model can be used to make predictions or arrive at an output. Once you have trained the model to some level of confidence you can use it to reason about data that it hasn't seen before and make predictions about the data.

Models In General

• https://gcmori.medium.com/what-are-the-top-nlp-language-models-71e4d780f5a0
• https://medium.com/@datasciencewsarah/5-models-every-data-scientist-should-know-33952ddf5bc9

How To Create A Model

• https://www.simplilearn.com/machine-learning-models-article

First Generation Models

• https://github.com/spinningideas/resources/wiki/AIML-%E2%80%90-Anthropic-%E2%80%90-Claude
• BERT - BERT (Bidirectional Encoder¹ Representations from Transformers)
• RoBERTa (Robustly Optimized BERT Pre-Training Approach)
• CodeBERT
• ALBERT
• OpenAI's GPT-3
• MUM - Multitask Unified Model

Current Frontier Models

This is a high-level, representative list of current frontier and frontier-adjacent models. “Frontier” is task-dependent and changes quickly, so use the linked provider documentation and live evaluations for the latest availability, capabilities, and rankings. The Agent Arena list is especially useful for comparing tool reliability, task completion, and steerability rather than general chat quality.

Major Proprietary Providers

• OpenAI — GPT-5.6 Sol and GPT-5.5, including high-effort reasoning variants highlighted in the Arena agent leaderboard. See the OpenAI API and OpenAI models documentation.
• Anthropic — Claude Fable 5, Claude Opus 5, and Claude Sonnet 5 are among the highest-ranked agent models in the current Arena snapshot. See the Claude model documentation and Claude 4 system card.
• Google — Gemini 3.6 Flash and Gemini 3.1 Pro, with Gemini 3.5 Flash and Gemini 2.5 Pro as additional reasoning and production options. See the Gemini API model catalog, Gemini API, and Gemma models.
• xAI — Grok 4.5, listed in the current Arena agent leaderboard. See the xAI documentation.

Open and International Frontier Models

• Moonshot AI — Kimi K3, including the Max agent variant ranked in the Arena agent leaderboard. See the Kimi platform documentation.
• Z.ai — GLM 5.2, including the Max variant ranked in the Arena agent leaderboard. See the Z.ai model documentation.
• DeepSeek — DeepSeek-V4 Pro and DeepSeek-V4 Flash, with local and fine-tuning tutorials available through Unsloth's model tutorials. See the DeepSeek API documentation.
• Alibaba Qwen — Qwen3.6, Qwen3.5, Qwen3-Coder-Next, and Qwen3-VL. These are listed in Unsloth's model tutorials; see the Qwen model collection.
• Mistral AI — Mistral 3.5, listed in Unsloth's model tutorials. See the Mistral models documentation.
• MiniMax — MiniMax M3 and MiniMax 2.7, listed in Unsloth's model tutorials. See the MiniMax open platform.
• NVIDIA — Nemotron 3 Ultra and Nemotron 3 Nano Omni, listed in Unsloth's model tutorials. See NVIDIA Nemotron.
• IBM — Granite 4.1, listed in Unsloth's model tutorials. See the Granite model family.

Evaluation and Local-Model Resources

• Arena Agent Leaderboard — live, community-driven evaluation of agentic model performance.
• Unsloth model tutorials — current tutorials for running and fine-tuning many open models locally.
• Unsloth model documentation — broader index of supported models and workflows.