What are the verified coding and reasoning benchmarks for Meta: Muse Spark 1.3?

Meta: Muse Spark 1.3 achieves SWE-bench Verified: 74.8% (Fast Thinking), LiveCodeBench: 66.2%. Verified sources: Artificial Analysis Intelligence Index v4.3.2 (Artificial Analysis), Muse Spark 1.3 High-Throughput Reasoning System (Meta AI).

How does Meta: Muse Spark 1.3 compare to rival frontier coding models on SWE-bench and pricing?

Meta: Muse Spark 1.3 achieves SWE-bench Verified: 74.8% at $1.2/$6 per 1M tokens. Direct peer comparison: OpenAI GPT-6 Sol (75.1% SWE-bench, $2.5/$12.5); xAI Grok 4.7 (73.6% SWE-bench, $3/$15); MiMo-V2.6-Pro (71.4% SWE-bench, $0.15/$0.6); Llama 4 Maverick (61.2% SWE-bench, $0.8/$2.4).

Does Swarm isolate Meta: Muse Spark 1.3 code edits from other agents?

Yes. Swarm provisions isolated Git worktrees for every session and subagent, preventing file race conditions and keeping your main working tree clean until verification passes.

Can Meta: Muse Spark 1.3 be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure Meta: Muse Spark 1.3 as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

Meta: Muse Spark 1.3 — Specs, Context & Capabilities

Q: What is the recommended Swarm multi-agent topology for Meta: Muse Spark 1.3?

Swarm recommends Meta: Muse Spark 1.3 as Coder Subagent (Tier 1: High-Speed Frontier Coder). Executes rapid code implementation inside worktrees, finishing 4x faster than slower planners. Best paired with: Claude Opus 5.5 or GPT-6 Astra for primary planning.

Q: How do I run Meta: Muse Spark 1.3 with Swarm?

Configure Meta: Muse Spark 1.3 via CLI using: swarm ctl model set --provider openrouter --model openrouter/meta/muse-spark-1.3, or switch models interactively inside the terminal TUI with /models.

Meta: Muse Spark 1.3 Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

74.8%

Fast ThinkingMeta AI Research ↗

LiveCodeBench

66.2%

ThinkingLiveCodeBench ↗

LMSYS Coding Elo

1420Elo

Coding ArenaLMSYS Arena ↗

Pricing ($ / 1M tokens)

$1.2/ $6

$0.12 cached readStandard API Pricing

What This Means in Real Software Development

Ties GPT-6 Sol at 48 on the Artificial Analysis Intelligence Index v4.3.2.
Generates 219 output tokens per second, making it 4x faster than typical frontier reasoning models.
Achieves 74.8% on SWE-bench Verified and 50 on Finance & Accounting Index.
Unmatched economic balance at $1.60 per Intelligence task with $1.20/M input pricing.

Swarm Multi-Agent Topology

Tier 1: High-Speed Frontier CoderRole: Coder Subagent

Best Paired With: Claude Opus 5.5 or GPT-6 Astra for primary planning

Executes rapid code implementation inside worktrees, finishing 4x faster than slower planners.

Git Worktree Isolation Advantage:

Fast generation allows 5 parallel Coder subagents to finish independent jobs in under 30 seconds.

What Developers & Practitioners Are Saying

Real-world field observations, production strengths, and developer consensus

Field Consensus

"A revelation in developer velocity: provides deep reasoning at near-streaming speed (219 TPS). Engineers report sub-minute completions for complex multi-file patches."

Verified Real-World Strengths

Incredible 219 TPS generation speed with frontier-class 48 Intelligence Index
Low $1.60 cost per task enables aggressive multi-attempt test loops
Strong AST comprehension and clean diff output

Production Caveats & Gaps

Slightly lower finance score (50) than Claude Opus 5.5 (61)
Prompt caching benefits require Meta infrastructure support

Primary Sources & Methodology Citations

Artificial Analysis Intelligence Index v4.3.2 — Artificial Analysis (September 2026)
Muse Spark 1.3 High-Throughput Reasoning System — Meta AI (June 2026)

Head-to-Head Frontier Comparisons: Meta: Muse Spark 1.3 vs Rivals

Standardized, side-by-side benchmark & economics comparison against peer models

Flagship Comparative Matrix

Model	Role / Tier	SWE-bench	LiveCodeBench	LMSYS Elo	AIME	Input / Output
Meta: Muse Spark 1.3OpenRouterCurrent	Tier 1: High-Speed Frontier Coder	74.8%	66.2%	1420	—	$1.2 / $6
OpenAI GPT-6 SolOpenAI Compare Meta: Muse Spark 1.3 vs OpenAI GPT-6 Sol →	Tier 1: Frontier Generalist	75.1%	65.8%	1415	—	$2.5 / $12.5
xAI Grok 4.7xAI Compare Meta: Muse Spark 1.3 vs xAI Grok 4.7 →	Tier 1: Algorithmic Reasoning Specialist	73.6%	64.9%	1408	—	$3 / $15
MiMo-V2.6-ProXiaomi Compare Meta: Muse Spark 1.3 vs MiMo-V2.6-Pro →	Tier 1: Maximum Efficiency Frontier	71.4%	63.5%	1395	—	$0.15 / $0.6
Llama 4 MaverickMeta Compare Meta: Muse Spark 1.3 vs Llama 4 Maverick →	Tier 2: Next-Gen Open Frontier Engine	61.2%	55.4%	1370	78.5%	$0.8 / $2.4

Explore All AI Model Comparisons Hub (Meta: Muse Spark 1.3 & Flagship Peers) →

Meta: Muse Spark 1.3 model overview

Meta: Muse Spark 1.3 is listed in the public Swarm catalog through OpenRouter. This page summarizes the verified catalog fields used by Swarm without claiming real-time access for every provider account.

Provider: OpenRouter
Family: OpenRouter
Context window: 1.05M tokens
Thinking support: low · medium · high · xhigh · max
Availability status: Available
Catalog ID: openrouter/meta/muse-spark-1.3

File input
Image input
Text input
Video input
Text output
Reasoning
Tools
Structured output
JSON

Run Meta: Muse Spark 1.3 in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider openrouter --model openrouter/meta/muse-spark-1.3

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign Meta: Muse Spark 1.3 to specialized system roles in ~/.swarm/config.json:

Primary / Planner: Orchestrates full task DAGs and multi-stage pipelines.
Coder Subagent: Generates code and unit tests inside private Git worktrees.
Finder Subagent: Performs read-only code search, symbol discovery, and mapping.

System Agent Roles →Why Swarm Architecture →