LLMMARA API pricing
5 MARA models. Source: https://cloud.mara.com/plans/pricing
- Provider type
- Inference host
- Server location (where inference runs)
- US
- Regions
- US
- Company HQ (secondary)
- US (outside EU) · Operated by MARA Holdings, Inc., a Nevada corporation, 1010 S Federal Hwy., Suite 2700, Hallandale Beach, FL 33009, USA, per the MARA Cloud Terms of Service and Privacy Policy. Checked 2026-11-03. · legal source
- EU data residency
- no
- Residency notes
- MARA Cloud Privacy Policy: data may be stored and processed in the United States and other countries outside the US. OpenRouter lists US datacenters. The public /v1/models API reports the SambaNova Cloud stack (owned_by sambanova).
- Residency source
- https://www.mara.com/legal/ai-policies
Primary location is where inference servers run (server region, from the cited source; otherwise 'unknown'). Company HQ is secondary public company information. OpenRouter prices are volatile: refreshed twice daily and excluded from cheapest-price answers.
Tap a box to filter by it; tap it again to remove that filter. Filters combine.
5 of 5 models.
| Provider | Model | Input | Output | Cached in | Blended 3:1* ▲ | Context | EU residency |
|---|---|---|---|---|---|---|---|
| MARA Inference host | gpt-oss-120b | $0.15 | $0.75 | — | $0.30 | 131K | no |
| MARA Inference host | gemma-4-31B-it | $0.38 | $1.15 | — | $0.573 | 131K | no |
| MARA Inference host | DeepSeek-V3.1 | $0.60 | $1.70 | — | $0.875 | 131K | no |
| MARA Inference host | MiniMax-M3 | $0.60 | $2.40 | — | $1.05 | 1.05M | no |
| MARA Inference host | DeepSeek-V3.2 | $3.00 | $4.50 | — | $3.38 | 33K | no |
Per 1M tokens, standard (non-promo) list prices; short-context / peak tier where tiered. USD unless shown with €, CHF or ₹: those are as published and not converted. ₹ rows also show ≈ USD at the RBI reference rate of ₹93.8579 per USD (2026-11-09); sorting still uses the published ₹ amount. Rows marked “Calculated from credits” have no official per-token price; see how they are calculated. Click a header to sort.
* Blended assumes 3 input tokens for every output token: (3×input + output) / 4. Your real mix will differ, so use the cost calculator with your own token counts for a real estimate.