Are these prices official?

They are editable estimates based on published list prices at the dates shown and are clearly labelled. Providers change pricing frequently, so confirm the current rate in each provider's dashboard before budgeting.

How do you decide which models are equivalent?

We group by capability tier — premium (flagship reasoning), balanced (mid-tier workhorse), and economy (small fast models) — so you compare like for like rather than by marketing name.

Why does the workload ratio change the delta?

Output tokens cost several times more than input tokens, and the two providers price that gap differently. A reasoning-heavy workload with lots of output can flip which provider is cheaper.

Is my data sent anywhere?

No. Everything runs in your browser. Nothing you enter is uploaded, stored, or logged.

Has LLM pricing gone up or down over time?

Broadly down. Intense competition between providers and efficiency gains have driven repeated price cuts at equivalent tiers, even as new flagship models launch at a premium.

Anthropic vs OpenAI Pricing Timeline

Anthropic vs OpenAI: how prices have moved

The two leading API providers price by capability tier, and those prices have shifted repeatedly as each releases new models and responds to the other. This tool lines up equivalent tiers — premium, balanced, and economy — and shows both the per-million-token history and the current delta for your workload.

How the comparison works

Each tier maps to comparable models from both providers. Costs are quoted in USD per million tokens, with input and output priced separately. The current delta blends those two rates using your workload ratio:

blended = (input_tokens / total) × input_price
        + (output_tokens / total) × output_price

Because output is far pricier than input, your input/output mix can change which provider comes out cheaper — so the default 1:1 ratio is only a starting point.

Tips for reading the timeline

Compare tiers, not names. A flagship from one provider should be compared with the other’s flagship, not its mid-tier model.
Mind the trend. Prices at a given tier have generally fallen; a model that looked expensive last year may have been cut since.
Use your real ratio. If your app generates long outputs, set the workload ratio accordingly — the headline list price understates output-heavy cost.