AI Models · Pricing

AI Model Pricing

Live per-token pricing for every major model — proprietary and open-weight — in one place. See costs at a glance in the chart, work out your monthly bill, compare models side by side, and find out what the same workload would cost to run on your own hardware: for open-weight models, nothing but electricity.

Pricing pulled from live data · tracked over time · olud.ai

Monthly cost calculatorEnter your token volume — costs update live and rank cheapest first.

Price at a glanceCheapest models by blended cost (input + output weighted 3:1). Open-weight in green.

Open-weightProprietary
Loading chart…

Bars scale to the most expensive model shown. Lower is cheaper.

Every model, every pricePer million tokens. Tick the boxes to compare models side by side. “Blended” weights input/output 3:1.

Compare

Loading live pricing…

API cost vs running it yourselfThe part nobody else shows: open-weight models cost ~$0 per token once you host them.

Every price above is what a provider charges to run a model for you. But open-weight models — DeepSeek, Llama, Qwen, Mistral, Gemma — are free to download and run on your own GPU. After the hardware, the marginal cost per token is basically electricity. For steady, high-volume workloads that flips the maths entirely.

For each popular paid model, here's the open-weight model people switch to, and where to read up on it:

Run open models yourselfFree is only free if you can actually run it. These are the tools to do it — full guides on each.

Ollama
One command to download and run open models locally. The easiest starting point.
Read the guide →
LM Studio
Desktop app to browse, download and chat with local models — zero terminal.
Read the guide →
vLLM
High-throughput serving engine for production self-hosting at scale.
Read the guide →
llama.cpp
Run quantised models on modest hardware — even CPU-only or a laptop.
Read the guide →
Jan
Open-source, offline ChatGPT-style app that runs models on your machine.
Read the guide →
Text Generation WebUI
Feature-rich web UI for loading, tuning and serving local LLMs.
Read the guide →
KoboldCpp
Single-file, GPU-accelerated runner — great for quick local inference.
Read the guide →
GPT4All
Private, local chat over your own documents, no internet required.
Read the guide →

Recent price movesProviders cut prices constantly. We log a snapshot daily and surface the changes.

Loading price history…

Keep exploring

Open-source LLM leaderboard →
Open-weight models ranked by real usage — the capability side of the story.
Full live model comparator →
Filter 300+ models by price, context, modality and provider.
Paid / proprietary models →
GPT-5, Claude, Gemini, Grok — pricing, context and a free alternative to each.
Model head-to-head comparisons →
Side-by-side breakdowns: open weights vs frontier, model vs model.
Open-source alternatives →
Self-hostable replacements for the paid AI tools you already use.
All tool guides →
Deep-dive guides on the open-source AI stack — what runs locally, and how.

Method. Prices are per million tokens, pulled from the OpenRouter API and refreshed daily. “Blended” cost weights input and output 3:1, a common single-number proxy. Self-host cost is marginal (electricity) and excludes hardware — actual figures depend on your GPU, utilisation and power price. Price-move history is logged from launch, so the moves section fills in over the coming weeks.

Want the capability side too?

Pricing tells you what a model costs. The leaderboard tells you whether it's any good — open-weight models ranked by real-world usage.

Open the leaderboard →
⚖ selected — Compare ↑