Processing 2,500 lines of code (~28,000 tokens) with Claude 3.5 Haiku costs $0.0224 for codebase ingestion and $0.0448 for an AI-powered code review and refactoring pass.
Quick answer
A 2,500 lines of code codebase is estimated at 28,000 tokens for Claude 3.5 Haiku. Ingestion costs $0.0224; a review and refactoring pass with roughly 20% output costs about $0.0448.
Method & trust
Code is estimated at a tokenizer-specific tokens-per-line ratio, then priced separately for repository context and generated review output. Comments, minified files, tests, and repeated context can materially change the bill.
Feed 2,500 lines of code into the prompt context for repository search, Q&A, or architecture planning.
Ingest 2,500 lines of code and generate audit findings, unit test recommendations, and refactor diffs.
Claude 3.5 Haiku writes 2,500 lines of code from scratch based on product specifications.
| Model | Provider | Code Ingestion | Cached Ingestion | Code Review Cost | Context Limit |
|---|---|---|---|---|---|
| Claude 3.5 Haiku (Current) | anthropic | $0.0224 | $0.00224 | $0.0448 | 200,000 |
| GPT-5.6 Luna | openai | $0.0051 | $0.00051 | $0.0112 | 1,050,000 |
| GPT-5.4 mini | openai | $0.0191 | $0.001912 | $0.0421 | 256,000 |
| GPT-5.4 nano | openai | $0.0051 | $0.00051 | $0.0115 | 128,000 |
| GPT-4o mini | openai | $0.003825 | $0.001912 | $0.006885 | 128,000 |
| Claude Haiku 4.5 | anthropic | $0.028 | $0.0028 | $0.056 | 1,000,000 |
On average, code yields approximately 11.2 tokens per line in Claude 3.5 Haiku (claude_bpe tokenizer). Indentation, brackets, camelCase variable names, and comments slightly increase token density compared to plain English text. 2,500 lines of code produces approximately 28,000 tokens.
Sending 2,500 lines of code as context and generating a thorough code review with recommendations costs approximately $0.0448. Utilizing prompt caching on repeat turns or static repository definitions drops this to $0.0246.
Claude 3.5 Haiku has a context window of 200,000 tokens. 2,500 lines of code consumes 14.00% of its total available context.