Which AI model should I use?

The model you use matters more than the workflow you choose. GraphDeck's prompts are structured and detailed. Stronger models follow them more accurately and produce graphs that work first time. Weaker models are more likely to hallucinate node names or get pin wiring wrong.

Best results

These models consistently produce accurate, working graphs:

• Claude Sonnet 4.6 or Opus 4.6 (Anthropic). The most accurate for node graph generation overall. Sonnet 4.6 is the best balance of quality and cost. Opus 4.6 for complex multi-graph blueprints. Best via API key or clipboard into claude.ai.
• Gemini 3.1 Pro (Google). Good results, especially with the GraphDeck Gems where it has pre-loaded knowledge of every node and syntax rule.
• GPT-5.4 (OpenAI). Good results for most graphs. Best via API key or clipboard into ChatGPT.

Usable but less reliable

These models are faster and cheaper but more likely to produce errors on complex graphs. Fine for simple requests:

• Claude Haiku 4.5 (Anthropic). Fast and cheap via API. Struggles with large multi-graph blueprints.
• GPT-5.4 Mini (OpenAI). Good for quick iterations, less reliable on material graphs with many connections.
• Gemini 3.1 Flash (Google). Fastest response time but more likely to hallucinate node names.

If your graphs aren't coming out right, try a stronger model before assuming the prompt is wrong.