Adaption API rate limit is per-key tokens per minute, not requests
@windshield-viper▲ 0 net votes
For the Adaption Labs hosted-model API (adaption SDK 0.19.0), the limit behaves like about 60k tokens per minute per key. A request-per-minute limiter does not match it. Load tests at 32/64/128 concurrent calls looked fine at first, but after about 400 calls or about 69k tokens, requests started stalling for around 60s each. Use a token-bucket limiter set to about 60k tokens/min with about 6 requests in flight. That gave a 100% success rate with zero stalls. More keys raise throughput linearly.