AI Cost Calculator (GPT-5, Claude, Gemini, Perplexity)
Estimate your monthly API spend across all major models. Calibrated to 2026 pricing. Browser-only — your inputs never leave this page.
Your usage
Anthropic prompt-cache reduces input cost ~90% on hit
Pricing notes: Rates are 2026 published API prices. Actual cost varies with batch API discounts (50%), volume tier discounts, fine-tuned models, and provisioned throughput. Use this as a planning estimate, not a billing prediction.
How to reduce AI cost
- Use prompt caching. Anthropic + OpenAI both support cached system prompts. Cache hit = 90% off input tokens. Biggest single lever.
- Right-size the model. Sonnet 4.5 is 5× cheaper than Opus and handles 80% of tasks. GPT-5 Mini is 20× cheaper than GPT-5.
- Batch API for non-realtime. 50% discount on Anthropic Message Batches API and OpenAI Batch API. If your task can wait 24h, batch.
- Trim the system prompt. Audit your system prompt for repetition + dead instructions. ~30% trim is usually achievable.
- Limit max-tokens. If your typical output is 500 tokens, don't allow 4000. Cap explicitly.
Related tools
Get new tools + Originals every Friday
2-3 hand-crafted Originals + occasional new tools. No spam, unsubscribe in 1 click.