Large Language Models
What is each model best at? At what price? At what privacy risk?
| Provider | Model | Open | Strengths |
|---|---|---|---|
| Anthropic | Claude 5 (Fable) / Opus 4.8 | AI coding, agentic work, creative reasoning | |
| DeepSeek | R1 V3 | ✓ | Scientific research, cost |
| ElevenLabs | Turbo v2.5 | Voice cloning, TTS | |
| Gemini 2.0 | Multimodal, deep research | ||
| Meta | Llama | ✓ | Open source ecosystem |
| Nous Research | — | ✓ | Decentralised models |
| OpenAI | GPT-4o / o3 | Broad ecosystem, agents | |
| Perplexity | — | Search-native reasoning | |
| Qwen | Qwen 2.5 | ✓ | Multilingual, open source |
| XAI | Grok | Real-time data, image gen |
Calibration stamp: 2026-07-16. Model tables are dated working theories — every row was verified against a moment in time, and rows rot silently as providers ship new generations (this table listed Claude 3.7 for months after it was superseded). Rows not re-verified at the stamp date may lag. Treat any unstamped model comparison, here or anywhere, as unverified.
For modality capabilities — voice, image, video — see AI Modalities.
The market commoditises fast. Model selection matters less than prompting capability — knowing how to direct whichever model you use. But "commoditised" does not mean "static": re-verify the model powering each critical job whenever a provider ships a new generation — an assumption calibrated on an old model is a silent cap on the new one (see AI Harness on unhobbling with proof).
This rapid turnover is Evolution made visible: new capability changes what can be attempted, while dated measurements stop yesterday's model assumptions from becoming today's invisible limits.
Model Selection
Review your Platform Stack regularly.
Focus on one critical job to be done at a time and master it.
- Identify a recurring need
- Find the best tool — cost, speed, accuracy
- Master functionality
- Glue to workflows
If the tool does not exist, investigate building it.
Subject Expertise
- Prompting — The capability that transfers across all models
- Text Prompts — Copy-paste templates and provider guides
- Visual Prompting — Image and video prompt structure
- AI Agents — AI agents and the jobs they perform
- AI Coding — Best AI coding tools and strategies
Context
- Evolution — Why changing capability requires dated reality checks and retained lessons
- AI Modalities — Voice, vision, video, audio, 3D capabilities
- Agent Frameworks — Platform for developing agents
- AI and Crypto — Decentralised compute and coordination
- SaaS Toolkit — Horizontal tools that amplify capabilities
Questions
Which model should you master — or does mastery of prompting make the model irrelevant?
- When does switching models cost more than learning to prompt the one you have?
- What job requires a specific model's strength that no other model can match today?
- How do you evaluate the privacy trade-off between open-source and closed-source models for your data?