indextkn

indextkn is a price index for AI models. It tracks the list prices of 985 models across 17 providers at the time of writing, shows them in one sortable table on the site, and serves the same data through a REST API, a hosted MCP server, a skill for coding agents, and webhooks that fire when a price changes. The pitch is that model pricing moves constantly and is scattered across provider pages, and nobody should be hand-maintaining a spreadsheet of it that's wrong by the time it's shared.

The problem is real for anyone who runs a product on several models or routes traffic between providers. The same model is often sold by more than one provider at different rates, prices have conditions like a higher rate past a prompt size or a cheaper off-peak window, and a change on a provider's pricing page doesn't announce itself. Cost estimates drift, and the drift only shows up on the invoice, by which point the routing decision that caused it is weeks old. It's also a data problem for tools. Cost calculators, router configs, and internal dashboards all embed prices somewhere, and every one of them goes stale independently.

The index covers text, image, video, speech, transcription, audio, embedding, and reranking models, with text making up the bulk at 704 entries. Providers include OpenAI, Anthropic, Google, Meta, xAI, DeepSeek, Mistral, OpenRouter, Moonshot AI, and Z.ai among others. Each entry shows input and output cost in US dollars per million tokens unless marked otherwise, the context window, and a confidence indicator. Green means verified, grey means partial with one price leg missing, amber means flagged with a check pending, and red means the price couldn't be confirmed, so you know how much to trust a number before it goes into a calculation. Conditional pricing is modelled explicitly, with step-up tiers past a prompt size and off-peak rates, and there's price history behind each offer. Clicking a model shows every provider that sells it side by side. The site notes that prices vary across providers for the same model, which is the whole reason the compare view exists.

For developers the API lives at api.indextkn.com under a v1 path with Bearer authentication and keys that start with itk_live_, which the docs say must stay server-side. The endpoints list models and their provider availability, return a single model across every provider, filter current prices by model, provider, or lab, compare the price spread for one model, and calculate the cost of a token workload across providers. There's a playground for trying calls without setup and llms.txt files for feeding the docs to an assistant. Webhooks watch one provider's price for one model and POST to your endpoint when it changes, with 30 days of delivery logs and a cap of 10 per account. The MCP server and skill let a coding agent look prices up directly while it's writing the code that calls the model. Filtering by lab as well as provider separates who trained the model from who sells access to it, a distinction the table keeps throughout.

It's for engineers who build on multiple models and need current costs in code, teams running routers or gateways that pick a provider by price, finance people estimating spend, and anyone writing a cost comparison who'd rather not scrape pricing pages. The changelog shows it launched on 28 August 2026 with 900 plus models and shipped webhooks, the MCP server, the skill, and tiered pricing support within days, so it's very new and moving fast. Because it's an index rather than a gateway, it never sits in the request path, so adopting it is a matter of reading a number, not routing traffic through a third party.

What sets it apart from static pricing tables and blog comparisons is that it's an API with verification status and change alerts, not a page. The confidence colours and the explicit modelling of tiers and off-peak rates are the details that matter when a number goes into a calculation, and the changelog records corrections to provider prices, which is at least evidence that someone is checking. The caveats are youth and scope. It tracks list prices rather than negotiated or usage-based discounts, the site doesn't publish its verification process in detail, and there's no named operator or contact, so if a price looks wrong there's no obvious place to report it.

Access is free. The API launched with 3,000 requests a month on a Bearer key, the dashboard manages keys, usage, and webhooks, and the site advertises free API access with a login. The pricing page returns a 404 and no paid tier is visible anywhere on the site, so treat the current posture as free with the possibility of plans arriving later. There's no published contact email, and the docs, changelog, terms, and privacy policy are the only supporting pages. The changelog is dated to the day and lists corrections alongside features, which is the closest thing to a status page the site has.