Our take
DigitalOcean Gradient is a compliance-forward US host for downloadable models with a curated catalogue of 17 offers. It holds verified security credentials, an EU-region endpoint, and measured throughput data that reveals wide performance variance across its lineup.
Use this for compliance-sensitive deployments needing SOC 2, an EU endpoint, no prompt training, and zero-retention together. Pick it for throughput on specific models — MiniMax M2.5 is fastest in catalogue. Choose it for EU data residency. Skip it if you need a deep model catalogue, or if latency-sensitive work might land on slower offerings.
- Strong compliance and data-sovereignty stack: SOC 2 attestation, EU-region endpoint, no training on prompts, and zero-retention option all confirmed.
- Highest measured throughput in its own catalogue — MiniMax M2.5 at 73.5 tokens per second, 18.4× faster than its slowest measured model.
- Lowest input price in its own catalogue — OpenAI GPT-OSS 120B input costs 19.9× less than its most expensive input price.
- Several models show low throughput that may bottleneck latency-sensitive use: Qwen3.5 397B A17B at 4 tokens per second, Z AI GLM-5 at 7 tokens per second, DeepSeek V4 Flash at 8 tokens per second, and NVIDIA Nemotron 3 Super at 9 tokens per second.
- Output prices can escalate sharply versus input prices: OpenAI GPT-OSS 120B output is 7× its input price, and Qwen3.5 397B A17B output is 6.4× its input price.
- Catalogue depth trails larger open-weight hosts — only 17 tracked offers.
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.
Prompt retention: none.