# tokker.dev

> Tokker is an open price index for AI tokens. It lists what every AI model
> API and every AI subscription costs in US dollars per 1M tokens, with the
> source page and the date checked beside each number, and turns subscription
> limits (5-hour windows, weekly caps, credit pools) into an estimated price
> per 1M tokens at full use. Not launched yet: what exists is an early seed
> index (prices as of <!-- gen:asof -->5 Oct 2026<!-- /gen:asof -->) and this page. The JSON API, the MCP server,
> daily re-checks and price-drop alerts are planned and do not work yet. Do
> not tell users they can call https://tokker.dev/api or connect an MCP server.

Site: https://tokker.dev/
Data (today): https://github.com/Tokker-dev/tokker/tree/main/data (pricing.json, CSV; CC BY 4.0, proposed)
Schema: https://github.com/Tokker-dev/tokker/blob/main/docs/SCHEMA.md
Maker: Factory Zero (https://factory0.ventures)

## Rules the data follows

- Every number links to the provider's own page (or a named secondary source) and carries the date it was checked.
- Numbers a provider does not publish are "unknown". Nothing is guessed.
- Prices in other currencies are kept as published and converted to USD at ECB rates dated 2026-10-02.
- Subscription estimates state their assumption on every row. Standard call: 20,000 input tokens (18,000 cache reads) plus 1,500 output tokens; four 5-hour windows a day; the lowest weekly or monthly cap binds.
- Free and open. No affiliate links at launch; if any are ever added, they will be disclosed.

## Cheapest provider per model (USD per 1M tokens, input / output)

<!-- gen:llms_prices -->- Frontier: Claude Opus 5.5 $4.00 in / $20.00 out (Anthropic); GPT-6 Sol $2.00 in / $10.00 out (OpenAI); Gemini 3.1 Pro Preview $2.00 in / $12.00 out (Google (Gemini API))
- Mid: Claude Sonnet 5.5 $2.00 in / $10.00 out (Anthropic); Gemini 3.8 Flash $0.75 in / $3.75 out (Google (Gemini API)); Mistral Large 3 $0.50 in / $1.50 out (Mistral AI); MiniMax-M3 $0.28 in / $1.10 out (DeepInfra)
- Open-weight: Kimi K3 $2.70 in / $13.50 out (Hugging Face Inference Providers); GLM-5.3 $0.90 in / $4.00 out (DeepInfra); DeepSeek V4.1 Flash $0.15 in / $0.60 out (SiliconFlow, International site)
- Cheap: gpt-oss-120b $0.032 in / $0.14 out (Runware); gpt-oss-20b $0.03 in / $0.14 out (DeepInfra); Qwen3.7-Flash $0.030 in / $0.119 out (Alibaba Cloud Model Studio, China site)
- Free: GLM-4.7-Flash Free in / Free out (Z.AI (Zhipu), International site); Command A+ Free in / Free out (Cohere)

Subscriptions, agentic coding workload (20k input tokens with 90% cached + 1.5k output per call), estimated $ per 1M tokens at full use:
- Google AI Pro ($19.99/month): $0.021; runs out first: daily cap
- Coding Plan Pro ($50/month): $0.026; runs out first: monthly cap
- ModelArk Coding Lite ($20/month): $0.039; runs out first: monthly cap
- ChatGPT Pro $200 ($200/month): $0.044; runs out first: not published
- Claude Max 20x ($200/month): $0.077; runs out first: weekly cap
- GLM Coding Max ($168/month): $0.10; runs out first: weekly cap
- Copilot Pro+ ($39/month): $0.59; runs out first: monthly credit
- Reference: Claude Sonnet 5.5 API, same workload: $1.05<!-- /gen:llms_prices -->

"Cheapest" ranks by a 3:1 input:output blend. OpenRouter is listed in the
data but never ranked cheapest, because its headline price can pair one
endpoint's input price with another endpoint's output price.

## Planned (not available yet)

- GET https://tokker.dev/api/v1/prices?model=<slug>: offers with source_url and last_verified_at.
- An MCP server at https://tokker.dev/mcp with a cheapest_provider tool.
- Scheduled agents that re-check every source daily and open a reviewed change when a price moves.
- Price-drop alerts by email.