New QuickSilver Pro models offer cost-optimized inference for long-context coding and agent tasks
Detailed pricing for Qwen 3.7 Plus and Kimi K2.7 Code, with a claimed ~20% cost savings over OpenRouter via routing optimizations.
Practical Summary
This post announces the availability of two new models on the QuickSilver Pro engine, with specific per-token pricing and a claim of being ~20% cheaper than OpenRouter due to routing optimizations. The models are positioned for heavy agent loops and long-horizon coding tasks, with integration via an OpenAI-compatible key.
Why It Matters
For teams optimizing AI tool spend, this provides a concrete new option with transparent pricing and a direct cost benchmark against a popular routing service. The focus on long-context windows (262K, 256K tokens) is relevant for complex coding and agentic workflows where context length impacts functionality and cost.
Understanding the New Model Offerings
QuickSilver Pro has launched inference for two models: Qwen 3.7 Plus and Kimi K2.7 Code. Both are designed for demanding tasks like persistent agent loops and complex, long-context coding projects.
Comparing the Pricing and Specifications
The post provides exact per-million-token costs: Qwen 3.7 Plus (262K context) is $0.256 for input and $1.024 for output. Kimi K2.7 Code (256K context) is $0.60 for input and $2.80 for output. The key claim is that routing optimizations make these models approximately 20% cheaper than using the same models via OpenRouter.
How to Evaluate and Integrate
To assess the cost saving, compare the listed per-token prices against your current OpenRouter rates for Qwen 3.7 Plus and Kimi K2.7 Code. For integration, the service uses a single OpenAI-compatible API key, allowing it to be dropped into existing stacks that support the OpenAI API format.