Archive · 3 Sep 2026
Meta Releases Muse Spark 1.3 in Advanced AI Race
Meta made a new computer program called Muse Spark 1.3.
It is designed to write code, automate tasks and follow complicated instructions better.
Developers can use it through Meta’s API.
Meta plans to add it to Instagram, Facebook and Meta AI.
The company says it uses 25% fewer tokens than the earlier version.
It can work on several workflows at once and ask questions when instructions are unclear.
Meta’s AI chief says it performs well against systems from OpenAI and Anthropic, although comparisons depend on the task.
Meta has not yet decided whether people will be able to download the model’s weights.
Meta says Muse Spark 1.3 improves coding, task automation and complex-instruction processing.
The model is available to developers through Meta’s API and will expand to Instagram, Facebook and Meta AI.
Alexandr Wang said it is competitive with Anthropic’s latest systems and beats OpenAI’s GPT-5.6 Sol on some coding challenges.
Meta says the model uses 25% fewer tokens than Muse Spark 1.2 and can manage multiple workflows simultaneously.
Meta has not decided whether to release Muse Spark 1.3’s weights, while its Watermelon model remains in development.
- Who
- Meta announced Muse Spark 1.3; Alexandr Wang discussed its performance.
- What
- Meta released a more powerful AI model focused on coding, automation and complex instructions.
- When
- The model became accessible to developers on Wednesday; no specific calendar date was given.
- Where
- It is available through Meta’s API and is planned for Instagram, Facebook and Meta AI.
- Why
- Meta says the model improves efficiency and performance while competing with leading AI systems from OpenAI and Anthropic.
Meta’s performance claims
Performance and access caveats
Competitive performance
Meta’s performance claims
Alexandr Wang said Muse Spark 1.3 is competitive with Anthropic’s latest systems, beats OpenAI’s GPT-5.6 Sol on some coding challenges and outperforms existing Chinese AI models.
Performance and access caveats
Wang cautioned that direct comparisons are difficult because performance varies by task and benchmarks may not reflect real-world use.
Model availability
Meta’s performance claims
Meta is making the model available through its API and plans to bring it to several Meta platforms at unchanged pricing.
Performance and access caveats
Meta has not decided whether to release the model weights, which means developers may not be able to download and independently adapt the model.
- Model
- Muse Spark 1.3, the successor to Muse Spark 1.2
- Availability
- Accessible to developers through Meta’s API
- Efficiency
- Requires 25% fewer tokens than Muse Spark 1.2, according to Meta
- Capabilities
- Improved coding, task automation, complex-instruction following and simultaneous workflow management
- Safety features
- Can request clarification and confirmation before potentially irreversible actions
- Pricing
- The same developer price as Muse Spark 1.2 through the Meta Model API
- Model weights
- Meta has not decided whether to release Muse Spark 1.3’s weights
- Next model
- The larger code-named Watermelon model is described as on track, but has no announced release date