Our take
Mara is a small, privacy-focused inference host with a narrow catalogue of five tracked offers. It emphasises prompt confidentiality and a zero-retention option over breadth or price leadership.
Use this when prompt privacy is paramount: it has an explicit no-training policy and a zero-retention option. Pick it for niche model access where its five tracked models match your need. Skip it if you need a deep catalogue, verified compliance attestations, or guaranteed EU data residency.
- Strong privacy guarantees relative to catalogue size: does not train on prompts and offers a zero-retention option.
- Fastest throughput in its own catalogue on minimax-minimax-m2-5 at 225.5 tokens per second, nearly 20% quicker than its next-best offer.
- Lowest input price in its own catalogue for openai-gpt-oss-120b, well under half the input price of its next-cheapest offer.
- Very limited catalogue depth: only five tracked offers.
- Severe throughput bottleneck on deepseek-deepseek-v3-1-terminus at 3 tokens per second, roughly 98% slower than its fastest offer.
- Compliance and geographic footprint are undocumented: SOC 2, headquarters country, and EU endpoint are all unverified in our data.
Point your tools here
We have not recorded what a router needs for Mara yet — no base URL and no docs link on file. That is our gap, not a sign Mara has no API; its own documentation is the place to look until we close it.
Privacy & data handling
These answers cover requests sent to Mara through OpenRouter, as OpenRouter records them, last read 58 min ago. For Mara's own API we hold no answer: check its terms. SOC 2 and the data processing agreement below are hand-checked, not part of that sync.
Prompt retention: none.