Meta Ships Muse Spark 1.3, Reporting Fewer Tool Calls per Task

Meta released Muse Spark 1.3 on 2 September 2026. Meta's launch post describes it as a multimodal reasoning model for long-running agentic, multi-agent and coding workflows, and reports that it uses roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 on the same tasks.

Meta's post shows a benchmark scorecard against Muse Spark 1.2, GPT-5.6 Sol and Claude Opus 5 across agentic, coding, instruction-following and long-context evaluations, but publishes the results as an image rather than as figures in the text, so no scores are quoted here.

The weights are closed. Meta lists an open-weights Muse Spark release on its roadmap without giving a date.

Meta does not publish per-token rates on the launch post. Artificial Analysis reports the standard endpoint at $1.25 per million input tokens and $4.25 per million output, with a 1M token context window. A separate Contributor endpoint is reported at roughly $0.10 and $0.20 per million tokens; that tier's condition is that the data is used to improve Meta's products, where the standard endpoint's is not. Some model trackers list the Contributor figures as the model's headline price without the condition.

On the Artificial Analysis Intelligence Index v4.1.1, Muse Spark 1.3 scores 62 at max effort.