What is Muse Spark 1.3 best suited for?
It is designed for long-horizon agent workflows, complex coding, codebase understanding, tool use, computer use, and multistep knowledge work.
MU
Muse Spark 1.3
Reasoning modelCoding modelMuse Spark 1.3 by Meta Superintelligence Labs
Release date: 2026-09-02Views: 8
Parameters
No data
Context length
1M
Multilingual
No data
Reasoning ability
No data
Muse Spark 1.3 is Meta's proprietary multimodal reasoning model released on September 2, 2026 for long-horizon agentic and coding workflows. It is available in Muse Code and Meta Model API with a 1M-token context window. Main-route pricing is $1.25 per million input tokens, $0.15 for cached input, and $4.25 for output. Max reasoning is pending additional safety testing; parameter count, maximum output, and knowledge cutoff are undisclosed.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Muse Spark 1.3
Model basics
Reasoning traces
Supported
Thinking modes
Thinking Level · Extra-High (Default)
Context length
1M tokens
Max output length
No data
Model type
Reasoning model
Modality (in / out)
Text, Image, Video → Text
Release date
2026-09-02
Model file size
No data
MoE architecture
No
Total params / Active params
No data / Not applicable
Knowledge cutoff
No data
Muse Spark 1.3
Open source & experience
Muse Spark 1.3
Official resources
DataLearnerAI blog
N/A
Muse Spark 1.3
API details
API speed
No data
💡Default unit: $/1M tokens. If vendors use other units, follow their published pricing.
Standard
| Type | Condition | Input | Output |
|---|---|---|---|
| Text | - | $1.25/ 1M | $4.25/ 1M |
Cache PricingPrompt Cache
| Type | TTL | Write | Read |
|---|---|---|---|
| Text | - | — | $0.150/ 1M |
“—” means the modality is not billed in that direction, or the vendor has not published a price for it.
Muse Spark 1.3
Benchmark Results
Muse Spark 1.3 currently shows benchmark results led by DeepSWE (1 / 34, score 75.40), Terminal-Bench 2.1 (2 / 49, score 88.80), AutomationBench (2 / 13, score 49.40). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
AI Agent - Tool Usage
3 evaluationsBenchmark / mode
Score
Rank/total
Muse Spark 1.3
Publisher
Facebook AI Research Lab
View publisher details Muse Spark 1.3 by Meta Superintelligence Labs
Model Overview
Official release
Muse Spark 1.3 is a proprietary multimodal reasoning model released by Meta Superintelligence Labs on September 2, 2026 under the API ID muse-spark-1.3. It is available in Muse Code and through Meta Model API, with a focus on long-horizon agents, complex instruction following, multitasking, reliable tool use, and coding efficiency.
Specifications and capabilities
Meta documents a 1M-token context window, text, image, video, and document input, and text output. The model can work across messy or conflicting sources, revise its plan, ask for help when needed, and use a real execution environment for visual reasoning over screenshots, clips, and documents. In internal engineering comparisons with 1.2, it used about 20% fewer tool calls and 25% fewer tokens. Previously available reasoning modes are live; max reasoning is still pending additional safety testing. Parameter count, maximum output, and knowledge cutoff are undisclosed.
Evaluations and safety
At max reasoning, Meta reports 1754 Elo on GDPval-AA v2, 64.9 on JobBench, 66.9 partial on OSWorld 2.0, 89.4 on DeepSearchQA, 49.4% on AutomationBench, 75.4% on DeepSWE v1.1, 59.4% on SWE-Atlas Codebase QnA, and 88.8% on Terminal-Bench 2.1. These are system-level results produced with benchmark-specific tools and harnesses. Meta also reports stronger resistance to adversarial input and prompt injection, plus better judgment around irreversible actions.
API pricing and data use
The main route costs $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens; its requests are not used to improve Meta products. The separate muse-spark-1.3-contributor route costs $0.10 / $0.002 / $0.20, but its data is used to improve Meta products. This catalog records only the main route's standard pricing. Muse Spark 1.3 remains closed-weight; Meta has announced only a future Muse Spark open-weights release.
Official sources
Muse Spark 1.3
FAQ
What is the context window?
Meta documents a 1M-token context window. It has not disclosed the maximum output length.
How much does the API cost?
The main route costs $1.25 per million input tokens, $0.15 per million cached input tokens, and $4.25 per million output tokens.
How does the Contributor route differ?
The Contributor route is much cheaper but its data is used to improve Meta products. It has a separate API ID from the main route.
Is Muse Spark 1.3 open weight?
No. Meta has mentioned a future Muse Spark open-weights release, but Muse Spark 1.3 is currently a proprietary hosted model.
DataLearner on WeChat
Follow DataLearner on WeChat for AI model updates and research notes.
