Claude Haiku WP Dev Benchmarks (vapvarun.com)

model-routing-wp-plugin-tasks.js

/**
 * Model routing for WordPress plugin tasks in Claude Code.
 * Routes tasks to the appropriate Claude model based on task type,
 * file count, and whether security judgment is required.
 *
 * Usage: import and call routeToModel(taskType, fileCount, requiresSecurity)
 * Returns: model ID string to pass as --model flag in Claude Code
 */

const MODELS = {
  haiku:  "claude-haiku-4-5",
  sonnet: "claude-sonnet-4-6",
  opus:   "claude-opus-4-7"
};

// Task types that Haiku handles at acceptable quality
const HAIKU_TASKS = new Set([
  "phpdoc",
  "wpcs_fix",
  "changelog",
  "readme_stub",
  "pot_file",
  "translation_strings",
  "unit_test_stubs",
  "simple_rest_endpoint",
  "admin_page_html",
  "boilerplate"
]);

// Task types that always need Sonnet minimum
const SONNET_TASKS = new Set([
  "dynamic_block_scaffold",
  "complex_rest_schema",
  "hook_timing_debug",
  "architecture_decision",
  "pr_review",
  "content_writing",
  "multi_file_refactor"
]);

/**
 * Route a plugin task to the right Claude model.
 *
 * @param {string}  taskType         - One of the task type keys above
 * @param {number}  fileCount        - Number of files involved in the task
 * @param {boolean} requiresSecurity - Whether the task involves security review
 * @returns {string} Model ID
 */
function routeToModel(taskType, fileCount = 1, requiresSecurity = false) {
  // Security review always gets Opus - cost is irrelevant here
  if (requiresSecurity) {
    return MODELS.opus;
  }

  // Multi-file tasks (5+ files) lose coherence on Haiku by file 5-6
  if (fileCount >= 5 && !HAIKU_TASKS.has(taskType)) {
    return MODELS.sonnet;
  }

  // Explicit Sonnet/Opus task types
  if (SONNET_TASKS.has(taskType)) {
    return MODELS.sonnet;
  }

  // Structured, low-ambiguity tasks: Haiku
  if (HAIKU_TASKS.has(taskType)) {
    return MODELS.haiku;
  }

  // Default: Sonnet for anything not explicitly classified
  return MODELS.sonnet;
}

/**
 * Estimate cost for a task before running it.
 * Based on median token counts observed across 300 Wbcom plugin tasks.
 *
 * @param {string} model    - Model ID
 * @param {number} inputK   - Estimated input tokens in thousands
 * @param {number} outputK  - Estimated output tokens in thousands
 * @returns {number} Estimated cost in USD
 */
function estimateCost(model, inputK, outputK) {
  const pricing = {
    [MODELS.haiku]:  { input: 0.00080, output: 0.00400 },
    [MODELS.sonnet]: { input: 0.00300, output: 0.01500 },
    [MODELS.opus]:   { input: 0.01500, output: 0.07500 }
  };

  const p = pricing[model];
  if (!p) throw new Error(`Unknown model: ${model}`);

  return (inputK * p.input) + (outputK * p.output);
}

// Example usage
const task = "phpdoc";
const model = routeToModel(task, 1, false);
const estimatedCost = estimateCost(model, 2, 1); // 2K input, 1K output

console.log(`Task: ${task}`);
console.log(`Model: ${model}`);
console.log(`Estimated cost: $${estimatedCost.toFixed(4)}`);
// Output:
// Task: phpdoc
// Model: claude-haiku-4-5
// Estimated cost: $0.0060

module.exports = { routeToModel, estimateCost, MODELS };