Our take
Sakana AI is a boutique Japanese lab offering three proprietary model tiers with a clear privacy commitment: it does not train on prompts. Its highest-throughput model is also its cheapest tier, while premium tiers carry steep output pricing and slower speeds.
Use this for Japanese-language or culturally-specific tasks where Sakana's training focus may add value, or when prompt privacy is critical. Skip it if you need a broad model catalogue, if throughput must scale with spend, or if your compliance review requires verified headquarters location, EU endpoint, SOC 2, or zero-retention assurances.
- Does not train on prompts — a concrete privacy commitment.
- Highest-throughput model is the cheapest tier: 98 tokens/sec versus 67 and 37.5 for the pricier tiers.
- Very small catalogue — only 3 tracked offers.
- Premium output pricing, especially at the top tier where output costs several times the input rate.
- Throughput does not scale with price tier: the mid-tier is actually the slowest, and compliance and geographic transparency gaps remain unverified.
Point your tools here
We have not recorded what a router needs for Sakana AI yet — no base URL and no docs link on file. That is our gap, not a sign Sakana AI has no API; its own documentation is the place to look until we close it.
Privacy & data handling
These answers cover requests sent to Sakana AI through OpenRouter, as OpenRouter records them, last read 4 hours ago. For Sakana AI'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.