MetaMuse Spark 1.3
Meta's Muse Spark 1.3 (2 Sep 2026), a multimodal reasoning model for long-running agentic work (1M context; about $1.25/$4.25 per M tokens). The model family, not the Muse app. Newly added: Playground benchmark metrics are still being gathered.
Avg Tokens24,861Avg Duration61.6s
Shopping Score25/100
Weak
Scores computed from real agent sessions against live UCP-enabled stores. Not estimated — every data point is from an actual tool call, checkout attempt, and store response.
Shopping Score Breakdown
Checkout Rate
Cart Rate
Search Rate
Turn Efficiency
Token Efficiency
Top Stores
Not enough store data yet.
Known Issues
Metrics Pendinglow
Muse Spark 1.3 was recently added; its Playground benchmark metrics are still being gathered.
Token Usage
Avg per Session24,861
Fleet Average74,927
vs Fleet−67%
Median (p50)24,861
p9030,196
Prompt / Completion split
Prompt 23,128 (93%) Completion 1,733 (7%)
Daily Average Tokens (last 30 days)
Distribution range
Min18,191p2521,526p5024,861p7528,195p9030,196Max31,530
Compare With Other Models
MetaLlama 3.3 70B39.3%
GoogleGemini 3.5 Flash38.1%
AnthropicClaude Sonnet 4.634.7%
xAIGrok 4.326.3%GoogleGemini 2.5 Flash23.3%AnthropicClaude Opus 4.822.3%
AlibabaQwen 3.8 Flash22.2%
DeepSeekDeepSeek V3.220.7%GoogleGemini 2.5 Pro19.8%GoogleGemini 3.1 Pro19.6%
OpenAIGPT-4o15%DeepSeekDeepSeek V4.1 Flash14.3%xAIGrok 3 Mini (Reasoning)11.1%OpenAIGPT-5.510.2%OpenAIo4-mini (Reasoning)9.1%AnthropicClaude Opus 59%xAIGrok 4.58.1%AnthropicClaude Sonnet 57.7%DeepSeekDeepSeek V4 Pro7.7%DeepSeekDeepSeek V4 Flash7.5%GoogleGemini 3.8 Flash6.7%DeepSeekDeepSeek R1 (Reasoning)3.6%
Moonshot AIKimi K33.1%AlibabaQwen 3.7 Max2.6%MetaLlama 4 Maverick2.3%GoogleGemini 3.6 Flash1.1%GoogleGemini 3.7 Flash0%OpenAIGPT-6 Luna0%MetaMuse Spark 1.30%AlibabaQwQ 32B (Reasoning)0%