Jaccard Similarity (Character Bigrams)

Choose the Jaccard bigram endpoint when you need a set-based similarity metric that is insensitive to word order and handles minor character-level differences well. It is ideal for near-duplicate detection, deduplication, and fuzzy entity matching where you care about shared character n-gram overlap rather than edit distance. Prefer this over Levenshtein when string length normalization matters less and over Dice when you want strict intersection-over-union semantics. It works well for short to medium-length strings like names, titles, and product labels.

• Missing required fields 'a' or 'b' returns a validation error
• Strings exceeding 20,000 characters are rejected
• Empty strings may produce undefined or 0 similarity due to empty bigram sets
• Non-string input types for 'a' or 'b' return schema validation errors
• Network or payment authorization failure returns 402 or 5xx status