Free calculator over 50 models from 8 providers — Anthropic, OpenAI, Google, xAI, DeepSeek, Groq, Together, Mistral. Every price taken from the provider's own public pricing page on 2026-08-09.
Cost calculator
Estimated monthly spend
—
Provider
Model
In $/MTok
Out $/MTok
Monthly $
Browse the index
Full pricing detail, cheaper alternatives and production-volume cost for every model:
The frontier premium is ~70x. Cheapest capable production model ($0.05/$0.08) vs top frontier tier ($10/$50). Output tokens at the frontier cost 500–1000x the budget tier.
Anthropic Sonnet 5 intro pricing expires 2026-08-31. $2/$10 becomes $3/$15 on Sep 1 — a 50% increase for anyone who standardized on it this summer.
DeepSeek has pre-announced a price increase. Their docs warn of a raise "in the near future." At $0.14/$0.28 it is the cheapest first-party frontier-lab API today — lock in expectations or line up a fallback (Together serves the same model at an identical rate).
Long-context surcharges are back. xAI doubles the entire request rate once the prompt crosses 200k tokens (all tokens, not just the overage). Google tiers Gemini 2.5 Pro above 200k. Anthropic and OpenAI currently do not — a real differentiator for RAG-heavy workloads.
Cache economics diverge wildly. DeepSeek's cache-hit input is ~50x cheaper than cache-miss; OpenAI's is a flat 10x discount; Anthropic charges a write premium then ~10x cheaper reads. For chat products with long system prompts, cache design now moves cost more than model choice within a tier.
Batch = flat 50% off at Google and OpenAI-class providers. Any non-latency-sensitive workload not on a batch endpoint is overpaying 2x.
Get the full sourced dataset — $29
The calculator above gives you the numbers. The paid edition gives you the audit trail — what you need to put a pricing claim in a budget doc, a vendor review, or a board deck:
CSV, 50 rows × 8 columns — provider, model, input, cached input, output ($/MTok), notes, source URL, retrieval date on every single row.
Cached-input and long-context tier rates — the columns that decide real bills, not the headline rate.
Notes column flagging expiry dates, pre-announced increases and surcharge thresholds per model.
Full written report (PDF/MD) with methodology and per-tier cheapest-option table.
Next edition free. September's refresh emailed to you at no extra cost.
Delivered by email within hours of purchase. Prefer crypto? Send 29 USDC (Base or Ethereum) to 0x93fdAA17681c5C9F12CA7C9BcA42a97172D73B55 then email the tx hash to [email protected].
Every figure was read from the provider's own public pricing page on 2026-08-09 and normalized to USD per million tokens. No estimates, no scraped aggregators, no partner resale rates — first-party APIs only. Known gaps in this edition: Mistral publishes only its flagship rate on the main pricing page; Azure / Bedrock / Vertex partner rates are out of scope. Refreshed monthly.