Cheapest LLM API — rank models by cost

The cheapest LLM API is a moving target. Blog posts with last quarter’s prices are how teams overpay. Use a live catalog sort, then check whether the cheap row is actually good enough for the job.

AI Benchmark Hub shows blended input/output cost per million tokens. Slide the cost weight to 100% on the leaderboard to re-rank by price, then compare the top cheap models against a quality leader so you see what you give up.

Cheapest models in the live catalog

Live OpenRouter pricing and Arena quality, cached about an hour. Not a static blog table.

ModelProviderBlend $ / 1MQualityContext
Mistral NemoMistral0.02131,072
Ling 3.0 FlashinclusionAI0.04262,144
Llama 3 8B LunarisSao10k0.048,192
MythoMax 13BGryphe0.064,096
Nex-N2-MiniNex Agi0.06262,144

Blended vs input vs output

A model can be cheapest on input and expensive on output. Chatbots with long answers care about output; RAG with huge prompts cares about input. The leaderboard blend is a starting rank. The compare table lists both. Cached input, when offered, can dominate if you send the same system prompt all day.

Free is not always cheapest

Free tiers 429, get deprioritized, or disappear. A $0.05/million row with an SLA is often cheaper in engineering time. See the free API guide.

Quality floors

Set a minimum quality index in your head (for example, only consider rows above 40) then sort by cost. Sorting only by price will surface models you will not ship.

Hosted open weights

Open-weight APIs are often the cheapest hosted options. Self-hosting can undercut them at high volume if you already run GPUs. Compare API price first; only then model the cluster.

How to read the live table

The table on this page is the five cheapest catalog rows with a positive blended price right now. Click a name for the monthly cost calculator at 1M / 10M / 100M tokens. That is the number to paste into a budget doc.

FAQ

What is the cheapest LLM API today?

See the live table on this page. It is pulled from OpenRouter and cached about an hour.

Are the cheapest models unusable?

Some are fine for classification and drafts. Compare quality index and run arena prompts before you standardize.

Does cheapest include GPT and Claude?

Their mini/haiku/flash SKUs sometimes appear near the top; flagships do not. Filter or search by name.

How is blended cost calculated?

Mean of input and output $ per million tokens when both exist. See methodology on /about.
cheapest LLM APIcheapest AI APIlowest cost LLMLLM pricing comparisonAI API cost per token
Rank by lowest API cost

Related guides

Also try: LLM leaderboard, compare tool, live arena.