Skip to content
Sovyron

LLM API prices and subscription plans, side by side

Compare LLM API prices per token for GPT, Claude, Gemini, DeepSeek, Qwen, Llama, Mistral and Grok across OpenRouter, OpenCode Zen, OrcaRouter, Requesty, Nano-GPT and the first-party labs — every rate normalized to USD per 1M tokens, with the cheapest offer highlighted for each model. Switch to subscription pricing to see the cheapest flat-rate coding plan per model, the usage level where it beats metered API billing, and which offers are free tier.

Models

151 of 151 models
Model

Browse all 659 comparable modelsOpen the comparison matrixCheapest models by capabilityRecent price changes

Quick answers

What is the cheapest LLM API per token?
As of Nov 22, 2026, the lowest output price among the 151 most widely offered models is $0.0185 per 1M tokens, for GLM-5.3-Flash on JoinGonka. The lowest input price is $0.0062 per 1M tokens, for GLM-5.3-Flash on JoinGonka. The cheapest LLM API ranking lists every model.
Are there free LLM APIs?
33 of the 151 compared models have at least one free-tier offer from a metered provider, for example Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, MiniMax-M3.1-Flash. The free LLM API list names the providers.
How current are the prices?
Metered API prices are refreshed once a day from provider catalogs and normalized to USD per 1M tokens. The last refresh was Nov 22, 2026. Flat-rate subscription fees are checked against vendor pricing pages every three days; see the methodology for how offers are ranked.
Is a subscription plan cheaper than the API?
It depends on how many tokens you use a month. Every plan has a break-even volume where its flat fee equals the metered API cost; the plans page lists it, and coding plan or API: which is cheaper shows how to work it out.
Why does the same model cost different amounts on different providers?
Each provider sets its own price and runs its own serving stack, so precision, context limits, rate limits and margin differ. Open-weights models show the widest gaps; why the same LLM costs different amounts walks through a worked example.
Does prompt caching make an LLM API cheaper?
It lowers the price of input tokens that repeat at the start of a prompt, and it leaves output tokens alone. Whether it pays depends on how often requests hit the cache; how much prompt caching saves shows the cached-input rates and a worked example.
Why do output tokens cost more than input tokens?
A model reads a prompt in one parallel pass and then writes the answer one token at a time, so providers price output higher. Output rates run several times the input rate; why output tokens cost more than input tokens shows the spread and what it does to a bill.