Pareto Code Router

Pareto Code Router is a text model from OpenRouter, distributed through api14 under the model id openrouter/pareto-code. It sits in the text category, so it is built for language work rather than image, audio or video generation. The name reflects how the model is put together: it is a router that maintains a tiered shortlist of strong coding models, ranked by Artificial Analysis coding percentiles. In other words, you are not calling a single fixed checkpoint but a routing layer that selects among coding-focused models according to their measured coding performance.

The model accepts text input and returns text output. Its context window is 2,000,000 tokens, which is large enough to hold entire repositories, long documentation sets or extended conversation histories in a single request. The provider description does not list a maximum output value, so treat output length as something to control through your request parameters rather than a documented fixed ceiling. Supported parameters include the usual sampling controls (temperature, top_p, top_k, min_p, top_a, repetition_penalty, frequency_penalty, presence_penalty), determinism aids (seed, logit_bias, logprobs, top_logprobs), and output shaping options (stop, max_tokens, max_completion_tokens, response_format, structured_outputs, verbosity). It also supports tool use through tools, tool_choice and parallel_tool_calls, and exposes reasoning controls via reasoning, reasoning_effort and include_reasoning.

Those capabilities point to a fairly specific set of use cases. The large context window suits codebase-level tasks: refactoring across many files, answering questions about a whole project, or generating changes that must stay consistent with existing code. Tool calling and structured outputs make it a reasonable fit for agentic coding workflows, where the model needs to invoke functions, read results and return machine-parseable payloads. The reasoning parameters are useful when you want the model to spend more effort on hard problems, and the seed plus sampling controls help when you need reproducible generations for testing or evaluation harnesses.

Calling it through api14 works like any other OpenAI-compatible model. Point your client at the api14 base URL, use your api14 key, and set the model field to openrouter/pareto-code. Because the surface is OpenAI-compatible, existing SDKs and libraries generally work without changes; you send a chat completions request with a messages array and add whichever of the supported parameters you need. The full parameter list above is available on the request, including the reasoning and tool-related fields, so you can tune behavior per call rather than being locked into defaults.

One practical consideration is routing itself. The provider description mentions a min_coding_score setting between 0 and 1, configured through the pareto-router plugin, which controls how the shortlist is filtered. If your task is not coding-related, or if you need a model with a documented maximum output length or non-text input modalities, another model may fit better. For text-based coding work where a very large context and tunable reasoning matter, this model is a sensible default to evaluate.