LLM pricing comparison: cost per million tokens

What this page compares

51 of the 51 chat variants in Quad Chat have a published provider price, from 18 labs. API prices are the model provider's list prices per million tokens, shown for comparison only. Quad Chat plans have a flat monthly price and draw usage from shared credits, so you are not billed per token. That distinction matters: a model that is cheap per token is not automatically cheap in Quad Chat, and an expensive one is not billed per token either. Cheaper models use fewer credits per message than frontier models, so credits go further on them.

API price rankings

Switch views to rank by blended cost, input, output or lab. Click a header to re-sort. Cheapest overall: gpt-oss 20B ($0.03 in / $0.14 out); Mercury 2.5 ($0.04 in / $0.15 out); Nemotron 3.5 Lightning 30B ($0.05 in / $0.20 out); DeepSeek V4 Flash ($0.076 in / $0.15 out); MiMo V2.6 Flash ($0.14 in / $0.28 out). Most expensive: GPT-5.5 Pro ($30.00 in / $180.00 out); Claude Fable 5 ($10.00 in / $50.00 out); GPT-6 Astra ($10.00 in / $50.00 out); GPT-5.5 Thinking ($5.00 in / $30.00 out); Muse Wonder ($4.00 in / $20.00 out). Cheapest input: gpt-oss 20B at $0.03, Mercury 2.5 at $0.04, Nemotron 3.5 Lightning 30B at $0.05. Cheapest output: gpt-oss 20B at $0.14, Mercury 2.5 at $0.15, DeepSeek V4 Flash at $0.15. A by-lab view lists each lab's price range.

What a typical request costs at list price

An example request with 2,000 input tokens and 500 output tokens (roughly a long prompt and a few paragraphs back), at provider list prices: gpt-oss 20B: about $0.0001. GPT-5.5 Pro: about $0.150, which is 1,154 times the cheapest. The lowest input price in the catalog is $0.03 per million on gpt-oss 20B; the highest output price is $180.00 per million on GPT-5.5 Pro. Reasoning models can generate extra thinking tokens that providers typically bill as output, so real costs for them run higher than a plain reply of the same length.

What Quad Chat actually bills

Quad Chat plans have a flat monthly price and shared credits instead of per-token billing. Web search is included and uses the same credits. Paying yearly is up to 15% less than paying monthly. Free: Monthly $0, Billed yearly $0, Credits / month 100, Chat variants 31 of 51. Go: Monthly $8, Billed yearly Monthly only, Credits / month 1,200, Chat variants 32 of 51. Pro: Monthly $15, Billed yearly $13/month ($156/year), Credits / month 2,000, Chat variants 46 of 51. Ultra: Monthly $40, Billed yearly $34/month ($408/year), Credits / month 5,000, Chat variants 50 of 51. Max: Monthly $200, Billed yearly $170/month ($2040/year), Credits / month 25,000, Chat variants 51 of 51.

How to use these numbers

Compare models within the same task: a cheaper model that needs two tries can cost more than a pricier one that gets it right. All 15 cheapest models run on the Free plan, so you can test them without paying. Sort by output price if you generate a lot of text or code, and by input price if you send long documents. For the full spec sheet behind each price, see the AI model comparison table or open any model's page.

Frequently asked questions

Which LLM is the cheapest per million tokens?

At provider list prices, gpt-oss 20B has the lowest blended price in the Quad Chat catalog ($0.03 in / $0.14 out per million tokens), and gpt-oss 20B has the lowest input price at $0.03. Prices change, so treat this as a snapshot of the catalog Quad Chat routes through.

What does blended price mean here?

It is an assumption, not a provider figure: three parts input price plus one part output price, divided by four. It approximates a chat where prompts are longer than replies. If your workload is mostly output, rank by the output column instead.

Does Quad Chat charge per token?

No. Quad Chat plans are a flat monthly price (Free $0, Go $8, Pro $15, Ultra $40, Max $200) with shared monthly credits. Cheaper models use fewer credits per message. The API prices on this page are the providers' list prices, shown for comparison.

Why are output tokens priced higher than input tokens?

Generating text costs providers more than reading it. In the catalog, 51 of 51 priced variants list a higher output price than input price, for example $0.03 in / $0.14 out on gpt-oss 20B. The table shows both so you can weigh them for your own workload.

Which labs have the widest price range?

Use the By lab view to see each lab's lowest and highest input and output prices. The catalog's overall range runs from $0.03 per million input tokens (gpt-oss 20B) to $180.00 per million output tokens (GPT-5.5 Pro).

Are the prices on this page up to date?

They come from the model catalog Quad Chat routes through and were last checked on October 5, 2026. Providers change prices, and Quad Chat updates the catalog when they do; the price shown to you in the model picker before you send a message is the one that applies.