Skip to main content

Model Selection Guide

Understanding SpecWeave's intelligent model routing

Overview​

SpecWeave uses a two-tier model strategy:

  1. Opus 4.6 - Default for all complex work (planning, analysis, architecture, code review)
  2. Haiku - For simple/cheap operations (translations, mechanical tasks)

The Models​

Opus 4.6 (Default - Planning, Analysis & Complex Work)​

Use for:

  • Strategic planning
  • Architecture design
  • Complex problem solving
  • Security analysis
  • Code review
  • Quality assessment

Pricing: $5 per 1M input tokens, $25 per 1M output tokens

Characteristics:

  • Deepest reasoning
  • Highest quality analysis
  • Best for complex tasks
  • Default for all agents

Haiku 4.5 (Simple & Cheap Operations)​

Use for:

  • Translations
  • Mechanical execution
  • Simple data processing
  • Configuration generation
  • Routine tasks

Pricing: $1 per 1M input tokens, $5 per 1M output tokens

Characteristics:

  • Fast execution
  • Cost-effective
  • Good for repetitive tasks
  • Used when task has detailed spec

Sonnet 4.6 (Balanced)​

Use for balanced speed and quality - Good middle ground between Opus and Haiku.

Pricing: $3 per 1M input tokens, $15 per 1M output tokens

Claude Code provides a hybrid model alias that works perfectly with SpecWeave's plan-mode-first workflow:

/model opusplan

This sets Opus 4.6 for plan mode (specs, architecture, analysis) and Sonnet 4.6 for execution (implementation, tests). Since SpecWeave mandates plan mode for all non-trivial work, you automatically get Opus reasoning where it matters most and Sonnet speed+savings during implementation.

See the Cost Optimization Guide for detailed savings estimates.

See full content at: https://github.com/anton-abyzov/specweave/blob/develop/.specweave/docs/public/guides/model-selection.md

Due to length, truncating here. File contains complete guide with agent classifications, phase detection algorithm, decision examples, troubleshooting, and FAQ.

Last updated: 2026-03-01 | SpecWeave v1.0.342