Meta released Muse Spark 1.3, an agentic coding model with efficiency improvements and new availability via Muse Code and the Meta Model API
Model release ● Confirmed 86% confidence first seen
Meta Superintelligence Labs launched Muse Spark 1.3, an agentic coding model aimed at long-horizon tasks. Coverage describes reported benchmark results and internal claims that it uses fewer tool calls and tokens than the previous version, with rollout through Muse Code and the Meta Model API (closed-weight, with additional participation-based pricing discussed).
Meta
Meta Superintelligence Labs
Muse Code
Muse Spark 1.3
Muse Spark AI coding model
Muse Voice Transcribe
Perplexity
0.6B classifier
Cartesia
Claude Code
DeepSWE
ElevenLabs
GPT-5.6-sol
Gemini 3.8 Flash
Google
MRCR 512K-1M
MSP protocol
MarkTechPost
Meta AI for Mac
Meta Model API
Meta Superintelligence
Muse Spark 1.2
OpenAI
Perplexity Computer
The New Stack
Decision brief
- What changed
- Meta Superintelligence Labs released Muse Spark 1.3, an agentic coding model for long-horizon software tasks, and made it available through Muse Code and the Meta Model API. Coverage says Meta reported internal efficiency gains versus Muse Spark 1.2, including about 20% fewer tool calls and about 25% fewer tokens, alongside new pricing options that can sharply discount API use when customers opt in to sharing prompts and outputs for training.
- Why it matters
- This gives leaders evaluating AI coding tools a new commercial option that combines model updates with broader product access, predictable Muse Code subscriptions, and API availability for integration. If Meta’s reported efficiency gains hold in production, the model could reduce inference and agent orchestration costs for coding workflows; however, the contributor-pricing structure also creates a governance tradeoff because the lowest prices require sharing usage data back to Meta for training.
- Evidence
- The event is supported across multiple outlets, with The New Stack and MarkTechPost both reporting the Spark 1.3 launch, availability in Muse Code and the Meta Model API, and Meta-stated benchmark or efficiency claims. TechCrunch independently corroborates the contributor-pricing approach and its data-sharing condition, while some additional claims in Latent Space/AINews about open weights are less consistently supported by the rest of the coverage.
- What remains uncertain
- Most performance and efficiency improvements cited are Meta-reported or based on selective benchmarks, so transfer to a company’s own repositories, agent flows, and cost profile is unverified. Coverage is also inconsistent on deployment posture: some reports say closed-weight availability today with no self-hosting, while others mention planned open-weight releases, so hosting flexibility and actual enterprise data exposure assumptions remain unresolved.
- Monitor next
- Watch for Meta to publish clearer enterprise terms on self-hosting or open-weight release timing, plus independent production benchmarks comparing Spark 1.3 cost and reliability on real coding workloads.
Analytical support, not advice — assumptions and open questions stated above.
Source coverage
Meta’s Claude Code rival exits beta with three new subscription tiers — and it’s pushing hard on price
The New Stack
Meta just beat OpenAI and Google at real-time transcription
The New Stack
Meta Superintelligence Labs Releases Muse Voice Transcribe: One Real-Time Model for Streaming ASR, Diarization, and Endpointing
MarkTechPost
[AINews] Muse Spark 1.3 matches GPT-5.6-Sol, confirming Meta Superintelligence as the newest Frontier Lab, >90% discount for training
Latent Space
newsletter
“Google was ahead only a few hours”: Muse Spark 1.3 edges out Gemini as Meta claims its biggest coding leap yet
The New Stack
Meta is paying to peek at how you use their latest AI model
TechCrunch
independent
Meta AI Released Muse Spark 1.3: An Agentic Coding Model That Uses ~20% Fewer Tool Calls and ~25% Fewer Tokens Than Muse Spark 1.2
MarkTechPost