Providers /Mara
Models
4
Regions
—
00
Editorial by LLMap · updated Aug 3, 2026Our take
Mara is a small, tightly-curated inference provider with an explicit zero-retention privacy promise and measured throughput on every tracked offer. Its catalogue is thin at just four models, and key compliance facts remain unverified in our data.
Use this when zero data retention is a hard requirement — this is a rare explicit guarantee. Pick it when throughput predictability matters, since every offer has measured tps. Skip it if you need a wide model choice, independently verified SOC 2, or a confirmed EU-region endpoint.
Strengths
- Explicit zero-retention option with no prompt training.
- Fastest measured throughput on minimax-m2-5 at 206 tps — more than five times the throughput of minimax-m2-7 and more than seven times that of deepseek-v3-1-terminus.
- Cheapest input price on openai-gpt-oss-120b relative to its own output, with input priced well below the minimax-m2 models.
Trade-offs
- Very small catalogue limits model choice — only four tracked offers.
- Compliance and geographic transparency gaps: SOC 2, headquarters country, and EU endpoint are all unverified in our data.
- Slowest throughput on the flagship-style model: deepseek-v3-1-terminus at 27.5 tps, which is more than seven times slower than minimax-m2-5 and more than five times slower than openai-gpt-oss-120b.
01
Privacy & data handling
- ✓
- Yes
- ✗
- No
- No answer on record
The colour says whether the answer favours you, not whether it is a yes — not training on your prompts earns a green cross, no zero-retention option earns a red one.
Trains on your prompts
✗No
Logs prompts
✗No
✓Offered
No answer on record
Prompt retention: none.
02