Free LLM‑pricing API launched

A language model bill is usually priced the way a phone plan used to be: one rate for what you send in, another for what comes back, plus special discounts for cached prompts and batch jobs. OpenAI’s pricing page, for example, lists GPT-5.4 at $2.50 per 1 million input tokens and $15.00 per 1 million output tokens. (openai.com) That sounds simple until you compare vendors, because each company publishes prices on its own page, in its own format, with its own exceptions. Google’s Gemini pricing page splits free and paid tiers, and Anthropic adds separate notes for prompt caching and batch processing on Claude pages. (ai.google.dev) (anthropic.com) So most teams still do model shopping by hand: open five tabs, copy numbers into a spreadsheet, and hope nothing changed last week. That is the gap Deep said it is trying to fill with a free language-model pricing site and application programming interface that went live this week and refreshes automatically with scheduled tasks. (x.com) The pitch is not that model prices are unknowable. The pitch is that prices move often enough that “official but scattered” becomes “hard to compare” the moment you need one clean table for a product decision. (openai.com) (ai.google.dev) (anthropic.com) That problem has gotten bigger as providers added more than one price per model. OpenAI now lists standard input, cached input, and output rates, while Google’s Gemini pages also separate free access from paid production usage and add search-grounding charges on some tiers. (openai.com) (ai.google.dev) Even one provider can have pricing rules that break a quick back-of-the-envelope estimate. OpenAI says GPT-5.4 sessions with prompts above 272,000 input tokens are charged at higher rates for the full session, which means “price per million tokens” is not always one flat number. (developers.openai.com) Google changed billing mechanics this month too. Its Gemini billing documentation says new prepay and postpay plans started taking effect on March 23, 2026, which is exactly the kind of operational change that makes stale comparison pages drift out of date fast. (ai.google.dev) That is why Deep’s mention of scheduled tasks matters more than the word “free.” A pricing directory that updates itself on a schedule is trying to behave less like a blog post and more like a weather report: useful only if today’s numbers are actually today’s numbers. (x.com) There is already a small industry of sites trying to solve the same mess. CostGoat, BenchLM, LM Market Cap, and Price Per Token all advertise live or recently updated language-model pricing tables, some covering 200 to 300-plus models. (costgoat.com) (benchlm.ai) (lmmarketcap.com) (pricepertoken.com) So the real story is not that one more comparison page exists. The story is that language-model pricing has become messy enough, across OpenAI, Anthropic, Google, DeepSeek, and others, that a free machine-readable pricing feed now looks like basic infrastructure instead of a side project. (openai.com) (anthropic.com) (ai.google.dev)