Grok 4.6 launches on OpenRouter with lower pricing and performance gains

Grok 4.6 from xAI is now available on OpenRouter and shows a clear improvement over Grok 4.5. Platform data indicate that the model outperforms other frontier models on the GPDVal-AA V2 benchmark, a standard evaluation that rates general capabilities, while keeping pricing at 2 dollars per million input tokens and 6 dollars per million output tokens (about 7.4 and 22 shekel respectively). OpenRouter notes that the average price customers actually pay is often lower than the official list price because of discounts and caching that reduce the effective cost.

OpenRouter does not host the model itself; it routes requests to multiple providers that run it. Three routing modes are offered: Balanced, which balances price and speed; Nitro, which prioritizes maximum speed; and Exacto, which aims for maximum accuracy in tool-calling. The system continuously monitors providers and automatically retries the next provider in line when one returns an error, resulting in a success rate measured over a 30-day window.

The platform publishes operational metrics that allow comparison of the model in a real-world setting: throughput, measured in tokens per second, where higher is better; latency, the round-trip time including processing, where lower is better; and TTFT (time-to-first-token), which measures the delay until the first token appears. These metrics are derived from actual traffic of public applications that send the largest request volumes to the model, providing a solid indication of production performance rather than laboratory results.

Standard evaluation scores are displayed as percent with a percentile rank showing where the model sits relative to all models on OpenRouter. Alongside the scores, the platform shows token volume and request counts over time as a signal of real adoption. For developers, calling the model uses an OpenAI-compatible API: swapping the base URL is sufficient, and the model slug is the only parameter that changes between models.