Providers /StreamLake
Inference provider

StreamLake

Models
20
Regions
00

Our take

Editorial by LLMap · updated Aug 2, 2026

StreamLake is a host for downloadable models with a mid-size catalogue and unusually high throughput on coding-specialist endpoints. It offers competitive pricing on select models, though operational transparency is sparse.

Use this for budget inference on mid-size models, or for high-throughput coding workloads where speed matters more than verified compliance. Pick it when you need the fastest coding-specialist throughput we have measured, or when the lowest input price for a mid-size model is the deciding factor. Skip it if your organisation requires verified SOC 2, a known headquarters location, an EU data-residency endpoint, or a zero-retention option.

Strengths
  • Lowest input price in sample for a mid-size model: Qwen3 30B A3B at 0.04815 per million input tokens, 2.7 times cheaper than DeepSeek V3-2 at 0.2145.
  • Highest throughput for coding-specialist models in sample: Kat Coder Air v2.5 at 130 tokens per second, 3.5 times faster than DeepSeek V4 Flash at 37 tokens per second and 6.2 times faster than DeepSeek V3-2 at 21 tokens per second.
  • Strong throughput on a second coding endpoint: MiniMax M2.5 at 106 tokens per second, 2.9 times faster than Qwen3 Coder Next at 63 tokens per second.
  • Does not train on prompts, confirmed by privacy policy.
Trade-offs
  • Operational transparency gaps: headquarters, EU endpoint, SOC 2, and zero-retention option are all unverified in our data.
  • Widest price spread within same model family: DeepSeek V4 Pro costs seven times more per input token than DeepSeek V4 Flash, with nearly identical throughput — 36 versus 37 tokens per second.
  • Several models priced above comparable alternatives in sample: DeepSeek V3-1 Terminus at 0.3426 per million input tokens is 3.8 times more expensive than Qwen3 30B A3B at 0.04815, with only 1.6 times the throughput — 60 versus 38 tokens per second.
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
Yes
No answer on record
No answer on record
02

Models & pricing

Something wrong on this page? Tell us