OpenAI: GPT-5.6 Luna

GPT-5.6 Luna is a text model from OpenAI, part of the GPT-5.6 series. OpenAI describes it as fast and cost-efficient, positioned for high-volume and latency-sensitive work rather than as the largest or most capable member of the family. It sits in the text category, so its purpose is language understanding and generation rather than audio, video or embedding tasks.

The model accepts text, images and files as input, and returns text only. Its context window is 1,050,000 tokens, which is large enough to hold very long documents, extended conversation histories or sizeable codebases in a single request. The maximum output is 128,000 tokens, so long generations are possible within that window. Its knowledge cutoff is 2026-02-16, and the provider release date is 2026-07-09. Because it handles image and file inputs alongside text, it can work over mixed material such as a document plus a screenshot plus a written question in one call.

Those characteristics point to a fairly specific set of uses. High-volume chat and classification are the obvious ones, since the model is built for throughput and low latency. Lightweight agentic workflows also fit: tool calling is supported, so the model can decide when to invoke functions you define and chain a few steps together. The large context window makes it reasonable for summarising or querying long reports, contracts or logs without chunking, and the image input allows questions about diagrams, screenshots or scanned pages. Reasoning controls mean you can dial effort up or down depending on whether a task needs deliberation or just a quick answer.

You call it through api14's OpenAI-compatible API using the model id "openai/gpt-5.6-luna". The supported request parameters are include_reasoning, max_completion_tokens, max_tokens, reasoning, reasoning_effort, response_format, seed, structured_outputs, tool_choice, tools and verbosity. In practice that means you can cap output length, request or suppress reasoning output, control how much reasoning effort is spent, ask for JSON or schema-conforming responses via response_format and structured_outputs, define tools and control tool selection, set a seed for more reproducible sampling, and adjust verbosity. If you already use an OpenAI-compatible client, pointing it at api14 and changing the model id is the main step.

There is no stated limitation in the facts beyond the text-only output and the knowledge cutoff, so treat those as the boundaries to check against your task. If you need output in a modality other than text, this model will not serve that need. If your workload is small and not latency-sensitive, the reasoning and verbosity controls still let you tune behaviour, but the model's stated design target is volume and speed. For work that depends on knowledge after 2026-02-16, plan to supply the relevant material in the prompt or through file and image inputs rather than relying on the model's internal knowledge.