Muse Spark 1.3

Tool callingStructured outputReasoningImage inputAudio inputVideo inputFile input

Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.

Context: 1.0M
Released: 1mo ago
Input / 1M tokens: $1.25
Output / 1M tokens: $4.25
Cached input / 1M tokens: $0.15

Sources:models.devOpenRouter

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

Other published benchmarks

Evaluation	Reported score and source
FrontierMath-Tier-4-v2-Private	46.3% Epoch AI · muse-spark-1.3_max Reasoning effort: max Epoch AI evaluation
FrontierMath-Tier-4-v2-Private	41.5% Epoch AI · muse-spark-1.3_xhigh Reasoning effort: xhigh Epoch AI evaluation
FrontierMath-Tiers-1-3-v2-Private	74.0% Epoch AI · muse-spark-1.3_max Reasoning effort: max Epoch AI evaluation
FrontierMath-Tiers-1-3-v2-Private	74.4% Epoch AI · muse-spark-1.3_xhigh Reasoning effort: xhigh Epoch AI evaluation
OTIS Mock AIME 2024-2025	99.2% Epoch AI · muse-spark-1.3_xhigh Reasoning effort: xhigh Epoch AI evaluation

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer: Meta
Model family: Muse
Type: Language model
Predecessor: —
Successor: —
Released: September 2, 2026
Model ID: meta/muse-spark-1.3
Openness: Closed
License: —
Input modalities: text, image, video, file
Output modalities: text
Modality: text+image+file+video->text
Tokenizer: Other
Max output tokens: 944K
Hugging Face: —
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Additional source evidence

Source model: meta/muse-spark-1.3
Source family: muse
Last updated: 2026-09-02
Model context: 1.0M
Model output limit: 131K
Open weights: No
File attachments: Yes
Reasoning: Yes
Tool calling: Yes
Structured output: Yes
Temperature control: Yes

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
GDPval-AA	1754 · Elo	Stirrup · max effort · 2 · 2026-09-02	Source
JobBench	64.9 · mean rubric score	OpenCode · max effort · 2026-09-02	Source
OSWorld	66.9 · mean partial score	Meta internal evaluation framework · 08.08 · max effort · 2.0 · 2026-09-02	Source
OSWorld	32 · binary completion rate	Meta internal evaluation framework · 08.08 · max effort · 2.0 · 2026-09-02	Source
DeepSearchQA	90.3 · F1	Meta browser harness · max effort · 2026-09-02	Source
Agentic IF Index	57.8 · score	Meta internal · max effort · 2026-09-02	Source
AutomationBench	49.6 · pass@1	public v3 task set · max effort · 2026-09-02	Source
MRCR	98.5 · mean sequence-match ratio	8-needle, 256K-512K · max effort · 2 · 2026-09-02	Source
MRCR	98.1 · mean sequence-match ratio	8-needle, 512K-1M · max effort · 2 · 2026-09-02	Source
DeepSWE	75.4 · pass@1	mini-swe-agent · max effort · 1.1 · 2026-09-02	Source
SWE-Atlas Codebase QnA	59.4 · pass@1	mini-swe-agent · max effort · 2026-09-02	Source
Terminal-Bench	88.8 · pass@1	Muse Code · max effort · 2.1 · 2026-09-02	Source

Complete source record (JSON)

{
  "id": "meta/muse-spark-1.3",
  "name": "Muse Spark 1.3",
  "description": "Muse Spark 1.3 is a multimodal reasoning model from Meta for long-running agentic, multi-agent, and coding workflows. It improves long-horizon agent collaboration, instruction following, and coding efficiency relative to Muse Spark 1.2.",
  "family": "muse",
  "last_updated": "2026-09-02",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "structured_output": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image",
      "video",
      "pdf",
      "audio"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 1048576,
    "output": 131072
  },
  "benchmarks": [
    {
      "name": "GDPval-AA",
      "score": 1754,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "Elo",
      "harness": "Stirrup",
      "variant": "max effort",
      "version": "2",
      "date": "2026-09-02"
    },
    {
      "name": "JobBench",
      "score": 64.9,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "mean rubric score",
      "harness": "OpenCode",
      "variant": "max effort",
      "date": "2026-09-02"
    },
    {
      "name": "OSWorld",
      "score": 66.9,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "mean partial score",
      "harness": "Meta internal evaluation framework",
      "variant": "max effort",
      "dataset": "08.08",
      "version": "2.0",
      "date": "2026-09-02"
    },
    {
      "name": "OSWorld",
      "score": 32,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "binary completion rate",
      "harness": "Meta internal evaluation framework",
      "variant": "max effort",
      "dataset": "08.08",
      "version": "2.0",
      "date": "2026-09-02"
    },
    {
      "name": "DeepSearchQA",
      "score": 90.3,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "F1",
      "harness": "Meta browser harness",
      "variant": "max effort",
      "date": "2026-09-02"
    },
    {
      "name": "Agentic IF Index",
      "score": 57.8,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "score",
      "variant": "max effort",
      "dataset": "Meta internal",
      "date": "2026-09-02"
    },
    {
      "name": "AutomationBench",
      "score": 49.6,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "pass@1",
      "variant": "max effort",
      "dataset": "public v3 task set",
      "date": "2026-09-02"
    },
    {
      "name": "MRCR",
      "score": 98.5,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "mean sequence-match ratio",
      "variant": "max effort",
      "dataset": "8-needle, 256K-512K",
      "version": "2",
      "date": "2026-09-02"
    },
    {
      "name": "MRCR",
      "score": 98.1,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "mean sequence-match ratio",
      "variant": "max effort",
      "dataset": "8-needle, 512K-1M",
      "version": "2",
      "date": "2026-09-02"
    },
    {
      "name": "DeepSWE",
      "score": 75.4,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "pass@1",
      "harness": "mini-swe-agent",
      "variant": "max effort",
      "version": "1.1",
      "date": "2026-09-02"
    },
    {
      "name": "SWE-Atlas Codebase QnA",
      "score": 59.4,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "pass@1",
      "harness": "mini-swe-agent",
      "variant": "max effort",
      "date": "2026-09-02"
    },
    {
      "name": "Terminal-Bench",
      "score": 88.8,
      "source": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "metric": "pass@1",
      "harness": "Muse Code",
      "variant": "max effort",
      "version": "2.1",
      "date": "2026-09-02"
    }
  ],
  "release_date": "2026-09-02",
  "open_weights": false
}

Infrastructure

Provider availability

All providers

Provider	Endpoint	Context	Input /M tokens	Output /M tokens	Cached input /M tokens	Cache write /M tokens	Uptime 1d	Data handling
Meta	meta	1.0M	$1.25	$4.25	$0.15	—	99.72%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15

Provider data:OpenRouter