LLM cost simulator for builders
Token counting is easy.
Real cost modeling is not.
Estimate the real cost of your LLM app before you ship. Model caching, batch, agent loops, tool calls, retries, and context growth across OpenAI, Claude, Gemini, and Llama models.
Frontier model pricing
All models →- Claude Opus 4.7 $15.00 / $75.00
- GPT-4o $2.50 / $10.00
- Gemini 2.5 Pro $1.25 / $10.00
- Llama 3.3 70B $0.59 / $0.79
Input then output, USD per 1M tokens. Cache, batch, and long-context tiers are on the individual model pages. Pricing is currently inferred from public docs; verify before relying on these figures for budgets. See changelog.
Quick token counter
Full counter →Cost estimate assumes 1,000 calls of this prompt as input plus 200 output tokens per call. Tokenizer for non-OpenAI models is approximated with cl100k_base or char/4, typically within 5 to 15% of true count. Pricing data last verified 2026-04-28.
What's different here
Most calculators stop at headline pricing. We model what actually drives the bill.
- 01
Caching ROI, not hand-waving
Anthropic's prompt cache cuts repeat input by up to 90%. Most calculators stop at the headline number. We model cache write cost against read cost against hit rate, and tell you the break-even point where caching actually pays.
Caching ROI - 02
Sync vs batch vs flex, decided
Batch is 50% off, but only fits async workloads. We frame this as a workload decision, not just a discount, so you know when to use batch and when not to.
Batch decision tool - 03
Agent loops, real cost
Multi-turn agents with tool calls, reasoning tokens, retries, and context growth cost very differently from single API calls. Project realistic monthly spend with workload templates for chatbots, coding agents, RAG, and batch classifiers.
Agent simulator - 04
Always current, transparently
Every model carries a last-verified stamp and an official-vs-inferred confidence flag. We publish a changelog. There's a "report incorrect price" link in the footer.
Changelog
Specific cost tools
Built around the questions builders actually Google.
More coming as the niche proves out — Cursor, Aider, Cline, agent-API workflows. One tool, one decision, no marketing fluff.
Ready when you are
Stop guessing. Model the bill.
The agent simulator handles the messy parts: turns, tools, reasoning tokens, retries, caching. Pick a workload template and see what the month actually costs.
Open the simulator