Muse Spark 1.3

Muse Spark 1.3 is positioned as a reasoning-oriented model from Meta, designed for complex, multi-step problem solving where accuracy is prioritized over raw speed. Independent coverage frames the release as targeting agentic coding workflows, with one third-party report highlighting efficiency gains of roughly twenty percent fewer tool calls and twenty-five percent fewer tokens than the prior Muse Spark 1.2 when completing the same coding tasks. A separately cataloged variant, Muse Spark 1.3 Contributor, is described as a reasoning model with vision, function calling, prompt caching, and structured-output support, fitting use cases that mix visual inputs with tool-driven, multi-turn execution.

In practical terms, the model is aimed at coding agents and other orchestrated pipelines where repeated tool invocations and long contexts drive cost and latency. The Contributor listing documents a one-million-token context window with up to roughly 944 thousand tokens of output, giving agent loops substantial room to maintain state across long sessions. The combination of vision understanding, structured response formats, and prompt caching makes it well suited for retrieval-augmented assistants, document- and image-grounded workflows, and developer tools that need reliable function calls rather than purely conversational responses.

Requestymuse-spark-1.3muse

Open provider page Provider docs

Quick Info

Provider: Requesty
Model key: muse-spark-1.3
Release date: Sep 2, 2026
Last updated: Sep 2, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.25
Output token cost: $4.25

Limits

Output tokens: 943,718 tokens
Context window: 1,048,576 tokens

Transparent token rates

Compare Muse Spark 1.3 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Muse Spark 1.3

Requesty

Coverage

Meta AI Research announced Muse Spark 1.3 on September 2, 2026, positioning it as an improved version of the Muse family for agentic and coding work. The post states the model handles longer-horizon tasks, juggles multiple workflows in a single thread, and proactively fills gaps in its own plans while tracking what it has learned across messy sources. The announcement says Muse Spark 1.3 asks clarifying questions on ambiguous prompts, confirms before consequential actions, and follows complex long-form instructions more reliably than earlier Muse Spark models. It also notes that a max reasoning level is available, and the model is reachable through Muse Code and the Meta Model API, with an evaluation report linked for benchmark detail.

Requesty

CoverageBenchmark

TrueFoundry's September 22, 2026 write-up reports that Meta Superintelligence Labs released Muse Spark 1.3 on September 2, 2026, about four weeks after version 1.2, calling it a natively multimodal model aimed at agentic and coding workloads. The article frames 1.3 as an efficiency-focused update rather than a raw intelligence jump, noting it uses roughly 20% fewer tool calls and 25% fewer tokens than 1.2 on comparable tasks. The same analysis lists Standard pricing at $1.25 per million input tokens and $4.25 per million output, with $0.15 cached reads and no long-context premium, and adds that 1.3 ships a max reasoning level and stronger prompt-injection resistance. The piece also reminds readers that Muse Spark weights remain closed, with an open-weights release still on Meta's roadmap.