The same model costs different prices on different platforms
Popular models are resold by many hosts. Grouping the listings reveals a price spread no single provider page will show you.
An AI model is not always a single product with a single price. Open-weight models are hosted by dozens of platforms, and even closed models are routed by third parties. Because every listing lands in this index, the same model name can appear several times at different rates — and this page is the only place that shows them side by side.
The spread, measured
Of the 942 distinct models in the index, 255 are sold by more than one platform we track. The widest spreads come from the most popular models, because hosting demand follows attention:
| Model | Listings tracked | Cheapest input /1M | Most expensive input /1M | Spread |
|---|---|---|---|---|
| GPT Oss 120b | 11 | $0.037 | $0.35 | 9.5× |
| GPT 4o | 9 | $2.5 | $5 | 2.0× |
| Glm 5.2 | 9 | $0.75 | $1.4 | 1.9× |
| Kimi K2.7 Code | 9 | $0.68 | $0.95 | 1.4× |
| Deepseek R1 | 8 | $0.4 | $5 | 12.5× |
| Deepseek V3 | 8 | $0.2 | $1.45 | 7.2× |
| Deepseek V4 Flash 0731 | 8 | $0.08 | $0.44 | 5.5× |
| Kimi K3 | 8 | $2.85 | $3 | 1.1× |
How to use this without getting burned
- Start from the canonical model page. It shows every listing for that model, which is the fastest way to see whether a cheaper host exists.
- Check what is bundled with the cheaper rate. Context window, cached pricing and capability flags differ between hosts serving the same weights.
- Verify the operational terms. Rate limits, data retention, region and support are not part of the token rate and are frequently the reason a platform is cheap.
- Re-check before you commit. These are list rates from a dataset that refreshes continuously; a host can change a price any time, and the changelog will show it.
Why we publish this instead of hiding the duplicates
Collapsing duplicate listings into a single canonical page is also the reason this site does not publish hundreds of near-identical pages. Each canonical page carries the price of every host for that model — original reporting that no single provider's pricing page can produce — while the duplicate listings remain reachable without being advertised to search engines.
Every provider in the index, with model counts and price floors.
See provider-level rankingsFrequently asked
› Why does the same model have different prices?
Some listings are the model owner's own API; others are platforms that host the same weights and set their own rate. Different hardware, margin and capacity assumptions produce different prices for identical output.
› Is the cheapest listing always the right choice?
No. Rate limits, throughput, data handling, uptime and latency are not in the price. Treat the spread as a shortlist, then verify the operational terms with the platform.
› How are listings grouped here?
By normalised model name, so variants of the same model collapse onto one canonical page that shows every platform's price for it. Duplicate platform listings stay reachable but are not offered to search engines.
Where the numbers come from
The data pages this article draws on — check any figure against them, they are recomputed on every refresh.