Meta has released Muse Spark 1.3 to developers, touting major gains in coding, complex instructions and autonomous AI tasks as competition with OpenAI and Anthropic intensifies.

Meta has launched Muse Spark 1.3, an upgraded version of its flagship artificial intelligence model, as the company steps up its competition with leading AI developers including OpenAI and Anthropic.

The new model is being made available to developers through the Meta Model API, with Meta retaining the same pricing structure used for Muse Spark 1.2. The company also plans to integrate the upgraded model into consumer products such as Facebook, Instagram and Meta AI, although it has not announced a specific timeline for that rollout.

Meta says the biggest improvements in the new model involve coding and agentic capabilities, allowing AI systems to complete sequences of tasks rather than simply answer individual prompts.

Meta Unveils New AI Model

Muse Spark 1.3 represents Meta's latest effort to strengthen its position in the increasingly competitive generative AI market.

Alexandr Wang, Meta's chief AI officer, described the release as the company's biggest improvement in model performance so far. He highlighted stronger performance on coding tasks and the model's ability to manage multi-step workflows.

The release comes as technology companies increasingly focus on AI agents capable of planning, reasoning and completing tasks with limited human intervention.

Bigger Focus On Coding

Coding is one of the main areas where Meta says Muse Spark 1.3 has improved.

The model is designed to handle more complicated programming tasks and maintain context across longer workflows. This could allow developers to use it for tasks ranging from generating and debugging code to managing multiple stages of software development.

Wang said the model performs better than OpenAI's GPT-5.6 Sol on coding benchmarks and is competitive with Anthropic's Claude Fable 5.1.

However, such comparisons should be treated cautiously because AI performance can vary significantly depending on the benchmark, task and evaluation methodology.

AI Agents Get Stronger

Meta is also putting greater emphasis on the model's agentic abilities.

Unlike conventional chatbots that respond to individual prompts, agentic AI systems can break down a larger objective into multiple steps and execute those tasks sequentially.

Muse Spark 1.3 is designed to manage several workflows at the same time and retain information over longer, more complicated assignments.

The improvement could make the model more useful for developers building autonomous assistants, coding tools and enterprise software capable of handling extended tasks.

25% Fewer Tokens

Meta claims Muse Spark 1.3 can complete tasks using about 25 per cent fewer tokens than the previous version.

Lower token consumption can potentially reduce operating costs and improve efficiency, particularly for developers running AI systems at scale.

The company also says the upgraded model follows detailed instructions more consistently and can handle complicated requests while maintaining relevant context.

These improvements are particularly important for AI agents, where a single task can involve many separate model interactions.

Meta Keeps Pricing Unchanged

Despite the reported improvements, Meta is keeping the pricing of Muse Spark 1.3 at the same level as Muse Spark 1.2 for developers using its API.

The decision could help Meta encourage wider adoption among developers and AI startups experimenting with its models.

The company is also expected to expand access through its own consumer ecosystem, including Meta AI, Facebook and Instagram.

Meta has not yet provided a firm date for when the upgraded model will become available across those services.

OpenAI And Anthropic Targeted

Meta's latest release comes as competition among major AI companies intensifies.

OpenAI and Anthropic continue to develop increasingly capable models focused on reasoning, coding and autonomous task completion. Meta is seeking to challenge those systems while also competing with rapidly improving Chinese AI models.

Wang claimed Muse Spark 1.3 currently performs better than Chinese AI models in several areas, while acknowledging that OpenAI is working on another advanced system.

Independent comparisons remain difficult because companies frequently use different benchmarks and testing conditions.

Watermelon Model Still Coming

Meta's AI ambitions extend beyond Muse Spark 1.3. The company is also developing a larger Watermelon model, which remains under development.

The new Spark release therefore appears to be an incremental step in Meta's broader strategy rather than the company's final answer to the latest generation of frontier AI systems.

For developers, the immediate focus will be on whether Muse Spark 1.3's reported coding, efficiency and agentic improvements translate into better real-world performance.

If the model delivers on Meta's claims, the release could strengthen the company's position in AI coding and autonomous-agent development while putting additional pressure on OpenAI and Anthropic.