Complete specs for every model on the gateway, with context limits, token pricing, cache rates, capabilities, and latency to set defaults and fallbacks.