toksum.dev

Predict your real LLM bill before shipping.

Compare Claude Opus 4.7, GPT-5.5, Gemini 2.5, and 40+ more — with batch discounts and prompt caching built in. Free. No sign-up required.

toksum.dev mascot — a friendly Sigma blob who keeps track of your LLM bill

Now tracking — 50+ models across 8 providers

OpenAI Anthropic Google xAI Meta DeepSeek Mistral Cohere Groq
See it in action

From your tokens to a decision report.

Count tokens, model your monthly workload, and get a one-page migration report — no signup, all in your browser.

Compare

Claude Opus 4.7 vs GPT-5.5

Top-frontier flagship comparison

Compare

Claude Haiku 4.5 vs GPT-5.4 mini

Budget tier for high-volume workloads

Tool

AI Token Counter

tiktoken o200k + Anthropic estimate — in your browser

Why toksum?

8 providers, 50+ models

OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, Cohere, Groq — frontier and budget tiers, side by side.

Real-world math

Batch API discounts, prompt caching ROI, output multipliers — modeled exactly the way they bill, not just a per-token sticker price.

Zero friction

All calculations run in your browser. No signup, no telemetry on inputs, no waiting on dashboards.

Migration reports

Free, fully-worked decision memos — one real workload, every credible option costed, the engineering tradeoffs spelled out.

All reports →
GO — but stay on OpenAI A RAG startup paying $360k/yr on GPT-5.5: should they switch? Cheap small model beats batching 5M classification calls a month: batch it, or switch models? Sonnet 4.6 if cost-driven OpenAI → Anthropic for a coding assistant: worth the migration?

Engineering guides

Deep, vendor-neutral writing on LLM cost — estimating a feature's bill, cutting output tokens, RAG architecture, prompt caching, and provider migration. Written for teams running real workloads, updated against official 2026 pricing.

All guides →
Guide · 11 min Estimate an AI feature's cost before you build it Guide · 10 min Cut LLM output token costs: 9 techniques that work Guide · 12 min RAG cost architecture: where the money actually goes Guide · 11 min Prompt caching ROI: when the 90% discount pays off

Built in public · Follow the journey on Twitter/X