$6–$13 per developer per active day. $150–$400/month for most professionals. Here's the complete per-token breakdown — and how to cut it by 40–70%.
Claude Code bills at Anthropic's API rates. Here's the current per-token cost for every model Claude Code uses.
| Model | Input ($/M tokens) | Output ($/M tokens) | Cache Read ($/M) | Cache Write ($/M) | Best For |
|---|---|---|---|---|---|
| Claude Sonnet 4.6 | $3.00 | $15.00 | $0.30 | $3.75 | Default — best balance |
| Claude Opus 4 | $15.00 | $75.00 | $1.50 | $18.75 | Complex reasoning tasks |
| Claude Haiku 4.5 | $0.80 | $4.00 | $0.08 | $1.00 | Simple tasks, high volume |
| Claude Sonnet 4 (prev) | $3.00 | $15.00 | $0.30 | $3.75 | Legacy sessions |
Cache reads are 90% cheaper than input tokens. Prompt caching is automatic for repeated context in multi-turn sessions.
Anthropic's published average is $6 per developer per active day. Here's what that means across different usage patterns — before optimization.
Most developers assume their prompts are the main cost. They're not. Here's what a typical 2-hour Claude Code session actually spends tokens on.
Six proven techniques — starting with the ones that have the highest impact per hour of effort.
Terse's 7-stage pipeline removes filler, fixes typos, strips redundancy, and compresses CLI output before it hits the context window. Average 40–70% reduction on prompts, 89% on CLI noise.
Highest ImpactPrompt caching reduces input token costs by 90% for cached portions. Set up a persistent system prompt with your project context, codebase overview, and preferences. Anthropic caches automatically for multi-turn sessions.
90% Savings on CacheA well-crafted CLAUDE.md file trains Claude Code to avoid redundant behavior from the first turn of every session. Terse generates CLAUDE.md rules from your session patterns automatically.
Session-Level SavingsHaiku 4.5 costs $0.80/M input vs. $3/M for Sonnet — nearly 4× cheaper. File edits, simple searches, and boilerplate tasks don't need Sonnet. Terse's Agent Monitor shows which turns are expensive so you can switch models deliberately.
4× Cheaper for Simple TasksIn a typical session, 34% of Read calls are redundant — Claude Code reads the same file multiple times. Each duplicate wastes the file's token count on input. Terse's Agent Monitor flags these in real time.
34% of Reads Are RedundantRun /compact at 200K–250K tokens to summarize context before it balloons. Claude Code's auto-compaction summarizes at the context limit, but proactive compaction at 60–70% capacity is cheaper and preserves more useful context.
Prevents Context BloatAnswers to the most common questions about Claude Code costs in 2026.
Terse reduces token costs across every major AI coding tool.
Terse compresses prompts, catches duplicate tool calls, and monitors per-turn cost. 30-day free trial — no credit card required until your trial ends.