Meta Enters Coding Agent Race with Muse Spark 1.1

Meta has released Muse Spark 1.1 in public preview, a multimodal model built specifically for agentic coding. It's the company's answer to Anthropic and OpenAI's dominance in developer tools built around autonomous AI agents.

What the new model can do

Muse Spark 1.1 is designed to diagnose and fix complex bugs, roll out new features in enterprise-grade systems, and handle large-scale code migrations. Meta emphasizes that the model supports the standard building blocks of modern agentic toolkits, including planning mode, goal conditioning, delegating tasks to subagents, and context compaction.

The million-token context limit lets the model actively manage its working memory over the course of a task. It's meant to recall earlier actions, pull information from much earlier stages of a project, and compact its history so that steps critical to ongoing work are preserved instead of keeping the entire raw conversation log.

Multimodality as a differentiator

Beyond coding, Muse Spark 1.1 is meant to excel at perception, multimodal reasoning, and using external tools. Meta points to generating code from an image, producing highly detailed descriptions of photos and video, and carrying out agentic tasks that combine different data types as key strengths. That's meant to set the model apart from competitors' text-only tools focused purely on code.

It's a strong agentic and coding model at a very low price, the strongest in agentic performance, tool use, and computer use - Mark Zuckerberg, CEO of Meta

Price as the main weapon

Meta openly acknowledges it still trails OpenAI and Anthropic somewhat in agentic model capability. So the company is leaning on price as its main selling point. At $1.25 per million input tokens and $4.25 per million output tokens, Muse Spark 1.1 lands close to Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna, though technically a bit above both models.

Zuckerberg also signaled that this won't be Meta's last announcement in this space in the near term. Muse Spark 1.1's launch coincided with an unusually intense week in the model market, one in which OpenAI shipped its GPT-5.6 family and xAI unveiled Grok 4.5, aimed specifically at programming tasks.

What it means for developers

For teams building their own agentic tools, another cheap, capable model adds further price pressure across the entire developer API market. Companies already using Claude Haiku or GPT-5.6 Luna gain an additional alternative with a similar price profile, which in practice strengthens their negotiating position with model providers.

For Polish software houses and startups building their own coding agents, it means an even wider pool of providers to test against specific tasks, from code migrations to multimodal work that combines screenshots with code generation.

Sources: TechCrunch (techcrunch.com), Dataconomy (dataconomy.com), eWeek (eweek.com), Meta AI blog (ai.meta.com)