Decision methodology
STACKOPTIMA turns provider specifications and normalized cost inputs into a workload-specific comparison. The output is a starting point for testing, not a substitute for your own benchmark.
Model pricing is compared per one million input and output tokens. Infrastructure is compared by the provider’s stated billing unit, region, and hardware class. Taxes, commitments, discounts, and data transfer may change the final bill.
Context window and provider-published capabilities are sourced directly. Intelligence and workload-fit indicators require a documented benchmark set and are never treated as provider facts.
Throughput and latency depend on region, load, prompt length, output length, and hosting route. They are comparable only when measured under the same test protocol.
Current reliability values are sample assumptions, not observed uptime. Production reliability requires an observed measurement window or published service evidence, with the window disclosed. A marketing uptime claim is not silently converted into an independent reliability score.