Technology
Meta launches Muse Spark 1.3 focused on agentic workflows and coding
The updated AI model offers a 1 million token context window, enhanced tool efficiency, and two API pricing tiers.
The short version
- Meta launched Muse Spark 1.3, an AI model designed for extended multi-step tasks, competitive coding, and native multimodal perception.
- The model is available through Muse Code and the Meta Model API with a 1 million token context window.
- Engineers claim the model consumes roughly 20% fewer tool calls and 25% fewer tokens than previous versions.
- Maximum reasoning modes are scheduled for release following the completion of additional safety evaluations.
Key facts
- Muse Spark 1.3 has rolled out in Muse Code and through the Meta Model API, featuring a 1 million token context window across variants.[Hacker News · Hacker News]
- The standard muse-spark-1.3 tier costs $1.25 per million input tokens, $0.15 for cached input, and $4.25 per million output tokens, with user inputs excluded from product training.[Hacker News]
- The muse-spark-1.3-contributor tier costs $0.10 per million input tokens, $0.002 for cached input, and $0.20 per million output tokens, with user data utilized for product improvements.[Hacker News]
- Meta engineers reported that the model operates with approximately 20% fewer tool calls and 25% fewer tokens compared to its predecessor.[Hacker News]
- The model features upgraded safety safeguards against prompt injections and improved handling of irreversible operations during multi-step tasks.[Hacker News]
What remains uncertain
- The precise launch date for maximum reasoning modes remains unannounced pending further safety assessments.[Hacker News]
- Timelines for future roadmap items, such as larger model variants and the open weights release, have not been specified.[Hacker News]
Sources
- Muse Spark 1.3Hacker News
- Introducing Muse Spark 1.3Hacker News