OpenAI: GPT-6 LunaActive

Create a ZenMux API key, set the OpenAI-compatible base URL to https://zenmux.ai/api/v1, and use openai/gpt-6-luna as the model ID. The ZenMux API quickstart

The token rate changes when a prompt exceeds 272K input tokens. At Standard rates below that threshold, input costs $0.10 per million tokens, cached input $0.01, cache writes $0.125, and output $0.50. Above it, the rates become $0.20, $0.02, $0.25, and $0.75, respectively. The higher rates apply to the full request, which explains the ranges on the ZenMux model page.

Yes. Repeated instructions, examples, or document context may qualify for a cache hit, which is billed at the lower cached-input rate. A hit is not guaranteed, so keep reusable content consistent across requests and inspect actual usage. See the ZenMux prompt cache guide

for prompt-layout advice and request logs

Start with Luna for a focused task, then test it against examples from your own workflow. If it misses your accuracy requirements on harder requests, you can escalate those requests to Sol. ZenMux’s model routing guide

Yes. GPT-6 Luna supports reasoning controls on ZenMux. Choose the configuration supported by your request protocol and evaluate results, response time, and token usage on your workload. ZenMux’s reasoning models guide

Yes. ZenMux supports tool calling for GPT-6 Luna. Your application defines the tools, executes any requested function, and returns the result to the model. Use the protocol and parameters supported by your endpoint, and validate model-specific tool behavior before production deployment.

GPT-6 Luna accepts image inputs and returns text, so it can analyze a supplied screenshot or other image. It does not return a generated image as its direct model output. Follow ZenMux’s multimodal input guide

to include images in a supported request.

Its 1,050,000-token context window describes how much context the model can handle, while its 128,000-token max output limits generated output. For document analysis or retrieval workflows, send material relevant to the question and account for the higher pricing tier when an input prompt exceeds 272K tokens.

For later news or changing information, provide current source material or connect an appropriate retrieval tool, then check the answer against that evidence. A standard model request should not be treated as an automatic live search.