Morph: Morph V3 Large

Morph: Morph V3 Large is a text model from Morph. It is presented as Morph's high accuracy apply model for complex code edits, and the provider describes it as reaching roughly 4,500 tokens per second with 98% accuracy for precise code transformations. That framing tells you what the model is built for: taking an instruction and an initial piece of code, then producing the edited code as output.

The model is text in, text out. It accepts text prompts and returns text completions, with a context window of 262,144 tokens and a maximum output of 131,072 tokens. That is a large working space, which matters when the code you are transforming does not fit in a small window. One important constraint comes from the provider: the prompt must follow a specific structure, with an instruction block and a code block, written as an instruction tag containing the instruction followed by a code tag containing the initial code.

Because the model is an apply model rather than a general chat model, the practical use cases follow from that shape. It fits automated code transformations, applying a described change to an existing file, refactoring or rewriting sections of code from a natural language instruction, and batch edits where the same kind of change has to be applied consistently across many inputs. The stated accuracy and throughput figures make it a candidate for pipelines where code edits are generated and applied programmatically rather than typed by hand.

You can call it through api14's OpenAI-compatible API using the model id "morph/morph-v3-large". The supported request parameters are logprobs, max_tokens, response_format, stop, structured_outputs, temperature, and top_logprobs. If you are porting existing code, that means the usual OpenAI style request body works, with the model field set to that id, and you can control sampling with temperature and stop, cap the output with max_tokens, and constrain the response shape with response_format or structured_outputs.

Keep the required prompt format in mind when you integrate it, since the instruction and code blocks are part of how the model is meant to be used. If your task is general conversation, open ended writing, or anything that is not a code edit, a general purpose text model may be a better fit than an apply model with a fixed prompt structure.