Meta Muse Spark 1.2: API Access and Evaluation Guide

Meta's Muse Spark 1.2 is listed as a coding-focused version of the Muse Spark family. Meta's developer pages now feature Muse Spark 1.3 as the current release, so older Spark 1.2 price, benchmark, and availability claims should be checked against the exact endpoint before reuse. This guide summarizes the provider-documented access path and gives a small evaluation plan without treating vendor claims as independent measurements.

Meta's developer landing page describes Muse Spark as available through the Meta Model API and Muse Code. It lists Spark 1.2 as a coding-optimized model and describes Spark 1.3 as the newer release. The page also links to an OpenAI-compatible API quickstart and OpenRouter. These are hosted access routes; the page does not publish a fixed price table for Spark 1.2 in the material linked here.

Meta's model/API page lists Spark 1.2 as part of the Muse family. Do not infer a parameter count, local-weight license, free quota, or stable endpoint name from the product description alone. Confirm those details in the current API reference, terms, and model catalog for the route you plan to use.

An earlier version of this article quoted benchmark scores, output speed, per-task cost, a one-million-token context limit, and an announced open-weights plan without a traceable primary source for each figure. Those numbers are removed here. A model being present in a provider catalog or aggregator does not verify its weights, quota, context limit, or training-data policy.

Use a small set of real repository tasks instead of relying on a broad label such as “frontier-level.” Pin the model ID and API route, keep the same code snapshot and tools, and define acceptance tests before running the model. Record task completion, test results, review corrections, latency, input/output usage, and rate-limit errors. Repeat tasks where randomness can affect the result.

If you compare Spark 1.2 with Spark 1.3 or another provider's model, do not mix different tool harnesses or coding-agent configurations and then attribute the full difference to the base model. Publish the model identifier, date, prompt, tool permissions, and whether each result came from your test or the provider's report.

The Meta Model API is the direct provider route; OpenRouter offers a separate aggregation path. Review the pricing, data handling, rate limits, and model identifiers for whichever route you select. If a Spark version is no longer offered by your route, update the application deliberately and rerun your own acceptance suite instead of assuming the replacement is behaviorally identical.