LLM Models Guide (Prompting Guide)

The Prompt Engineering Guide includes model-specific prompting guidance for 21+ LLMs, from foundational models to the latest releases.

Major Model Families (confidence: high)

OpenAI

• ChatGPT (GPT-3.5-turbo) — Conversational RLHF-tuned model, ChatML format, multi-turn prompting
• GPT-4 — Multimodal (text+image), top-10% bar exam, steerable via system messages, vision capabilities
• Sora — Video generation model (separate from text LLMs)

Anthropic

• Claude 3 — Haiku/Sonnet/Opus family, 200K context window, outperforms GPT-4 on several benchmarks, vision capabilities

Google DeepMind

• Gemini — Family spanning Ultra/Pro/Nano sizes, multimodal by design
• Gemini Advanced / 1.5 Pro — Extended capabilities and context windows
• Gemma — Open 2B/7B models, prompt format with control tokens

Meta

• LLaMA — Foundational 7B-65B, LLaMA-13B > GPT-3 (175B)
• LLaMA 3 — 8B/70B/400B, 128K vocabulary, 15T tokens, GQA, SFT+PPO+DPO alignment
• Code Llama — Code-specialized LLaMA variant

Mistral AI

• Mistral 7B — Efficient 7B model competitive with larger models
• Mistral Large — 81.2% MMLU, 32K context, multilingual (FR/DE/ES/IT), function calling
• Mixtral 8x7B — MoE model (12.9B active/46.7B total)
• Mixtral 8x22B — MoE (39B active/141B total), 64K context, Apache 2.0, 90% GSM8K

xAI

• Grok-1 — 314B MoE, Apache 2.0, 63.2% HumanEval, 73% MMLU

Other Notable Models

• Flan — Google’s instruction-tuned T5/PaLM (1.8K tasks), strong zero-shot/CoT
• Phi-2 — Microsoft’s small language model
• OLMo — AI2’s open-source language model
• Kimi K2.5 — Moonshot AI’s 1T-param MoE, 96.1% AIME, agent swarm orchestration

Full Collection (confidence: high)

The Model Collection page catalogs 47+ notable LLMs from BERT (2018) through PaLM 2 (2023), providing historical context for model evolution.

Implications

The breadth of models covered (from 2022’s Flan to 2025’s Kimi K2.5) shows how rapidly the LLM landscape has evolved. The guide tracks key trends: MoE architectures, extended context windows, multimodal capabilities, and the shift toward open-weight models.

prompt-engineering-guide | dair-ai | flan | chatgpt | llama | gpt-4 | mistral-7b | gemini | phi-2 | mixtral | code-llama | olmo | sora | claude-3 | gemma | grok-1 | kimi-k2.5 | llama-3 | mistral-large | mixtral-8x22b | model-collection

Sources

• raw/prompts/articles/promptingguide-pt-models.md
• raw/prompts/articles/promptingguide-pt-models-chatgpt.md
• raw/prompts/articles/promptingguide-pt-models-gpt-4.md
• raw/prompts/articles/promptingguide-pt-models-claude-3.md
• raw/prompts/articles/promptingguide-pt-models-gemini.md
• raw/prompts/articles/promptingguide-pt-models-llama.md
• raw/prompts/articles/promptingguide-pt-models-llama-3.md
• raw/prompts/articles/promptingguide-pt-models-mistral-7b.md
• raw/prompts/articles/promptingguide-pt-models-mistral-large.md
• raw/prompts/articles/promptingguide-pt-models-mixtral.md
• raw/prompts/articles/promptingguide-pt-models-mixtral-8x22b.md
• raw/prompts/articles/promptingguide-pt-models-grok-1.md
• raw/prompts/articles/promptingguide-pt-models-gemma.md
• raw/prompts/articles/promptingguide-pt-models-kimi-k2.5.md
• raw/prompts/articles/promptingguide-pt-models-flan.md
• raw/prompts/articles/promptingguide-pt-models-phi-2.md
• raw/prompts/articles/promptingguide-pt-models-code-llama.md
• raw/prompts/articles/promptingguide-pt-models-olmo.md
• raw/prompts/articles/promptingguide-pt-models-sora.md
• raw/prompts/articles/promptingguide-pt-models-collection.md