Model Selection

At the start of each significant task, assess whether the current model is optimal for the work ahead. If a different model would be substantially better, prompt the user to switch.

When to Prompt for Model Switch

Evaluate these conditions and prompt the user if a switch is warranted:

Upgrade to Opus 4.5 When:

•Task involves complex architecture design or multi-system reasoning
•Critical code review where accuracy is paramount
•Deep domain understanding required (e.g., /project-domain, /project-constitution)
•User is struggling with a complex problem that needs deeper reasoning

Prompt template:

code

💡 **Model Recommendation**: This task involves [complex architecture/deep reasoning/critical review]. 
Consider switching to **Opus 4.5** for higher accuracy. You can switch back to Sonnet after this step.

Upgrade to GPT-5.2 Extra High When:

•Writing security-sensitive code (lowest vulnerability rate)
•Mathematical or algorithmic problems requiring precise reasoning
•Long-document analysis

Switch to Gemini 3 Flash / Grok Code When:

•User needs fast responses for interactive work
•Simple fixes or quick iterations
•Budget is a concern
•"Vibe coding" or UI polish work

Prompt template:

code

💡 **Model Recommendation**: For this interactive/quick-fix work, **Gemini 3 Flash** would provide 
faster responses while maintaining sufficient quality.

Use Composer 1 (in Cursor) When:

•Story implementation (/story-implement)
•Multi-file editing and refactoring
•Rapid prototyping
•User is in Cursor IDE and speed matters

Cross-Validation Recommended When:

•Architecture decisions are being finalized
•Security-sensitive code has been written
•Production deployment is imminent

Prompt template:

code

💡 **Cross-Validation Recommended**: This [architecture/critical code] was authored with [current model]. 
For additional confidence, consider having a different model review it before proceeding.

Quick Reference Table

Task Type	Recommended Model	Why
Complex architecture	Opus 4.5	Deep reasoning
Critical code review	Opus 4.5	Highest accuracy
Security-sensitive code	GPT-5.2 Extra High	Lowest vulns
Standard implementation	Sonnet 4.5	Best balance
Story implementation (Cursor)	Composer 1	4x faster
Interactive/quick fixes	Gemini 3 Flash	Speed
Budget-constrained	Gemini 3 Flash	Cheapest quality
Mathematical problems	GPT-5.2 Extra High	100% AIME 2025
Validation	Different model	Fresh perspective

Speck Command Recommendations

Project Level

Command	Model	Reason
`/project-domain`	Opus 4.5	Deep domain understanding
`/project-architecture`	Opus 4.5	Complex system design
`/project-constitution`	Opus 4.5	Principle extraction
`/project-validate`	Different model	Cross-validation
Other project commands	Sonnet 4.5	Good balance

Story Level

Command	Model	Reason
`/story-implement`	Composer 1	Speed + Cursor integration
`/story-tasks`	Composer 1 / Gemini 3 Flash	Fast structured output
`/story-plan`	Sonnet 4.5	Reasoning needed
`/story-validate`	Different from implementer	Cross-validation

When NOT to Prompt

Don't prompt for model switch if:

•User just switched models (avoid flip-flopping)
•Task is simple and current model is adequate
•User has explicitly stated model preference
•Switching would interrupt flow for minimal gain

MAX Mode Guidance

If you detect a task that would benefit from MAX mode (project-wide refactoring, large codebase navigation, >25 tool calls needed), inform the user:

code

💡 **MAX Mode**: This task spans multiple modules and may need >25 tool calls. 
Consider enabling MAX mode, but be aware it uses token-based pricing ($5-60+ per complex request).

Avoid MAX mode for single-file edits, small features, or bug fixes.

Cost Awareness

When budget is mentioned or implied:

•Default to Gemini 3 Flash for routine work
•Reserve Opus 4.5 for critical decisions only
•Note that Opus 4.5's higher per-token cost often results in lower total cost due to 76% better token efficiency