Meta releases Muse Code in beta, a terminal coding agent powered by Muse Spark 1.2, a coding-focused model priced at $1.25/1M input and $4.25/1M output tokens (Jonathan Vanian/CNBC)

Meta has unveiled Muse Code, its own coding assistant leveraging a fresh AI model named Muse Spark 1.2. This tool aims to challenge rivals like Anthropic's Claude Code and OpenAI's Codex. Muse Code operates through a terminal interface and is built on improvements in code generation, debugging, understanding of codebases, and overall developer workflow.

The AI model can autonomously handle software engineering tasks such as writing code, planning modifications, and validating results. This allows users to create functional software using straightforward text prompts. Moreover, Muse Code can orchestrate multiple sub-agents to delegate tasks, streamlining the overall process.

Meta showcased several applications for Muse Code, including an interactive model of a photon sphere and a homage to Plants vs. Zombies. They also demoed the tool generating a webpage from an MP4 file. In terms of pricing, Meta seeks to position Muse Code as more affordable than its competitors. By default, the tool follows the same pay-as-you-go model as Muse Spark, charging $1.25 per million input tokens and $4.25 per million output tokens.

Additionally, Meta plans to offer a discounted "contributor tier" for users willing to provide feedback to enhance the coding agent. This lower-cost tier would be priced at $0.10 per million input tokens and $0.20 per million output tokens. In comparison, Anthropic's Sonnet 5 model typically costs $3 per million input tokens and $15 per million output tokens.

If Muse Code performs as expected, this pricing advantage could sway companies to prefer Meta's offering over those of its competitors.