<>
Meta has released its Muse Code programming agent in public beta alongside version 1.2 of the Muse Spark model, with pricing closer to Chinese AI model costs than to standard rates charged by Western competitors, according to market analysis.
Pricing Strategy and Usage Tiers
According to Meta, the standard usage pricing for Muse Spark 1.2 is set at $1.25 per million input tokens and $4.25 per million output tokens, with cached input priced at $0.15 per million tokens. The company also offers a “contributor” mode that lowers costs further to $0.10 per million input tokens, $0.20 per million output tokens, and $0.002 per million cached tokens. In exchange for the reduced rates, the contributor mode allows Meta to utilize user data to refine its language models and enforces usage limits on a rolling five-hour window rather than a strict request count, as detailed in company documentation.
Enterprise Data Security Controls
Meta has introduced zero data retention requests for enterprise customers using the platform. The pricing model positions Meta closer to Chinese AI model costs than to standard rates charged by Western competitors, according to market analysis. By comparison, competitor pricing includes Claude Sonnet 5 at $2.00 per million input tokens and $10.00 per million output tokens, while GPT-5.6 Terra costs $2.50 per million input tokens and $15.00 per million output tokens.
Model Performance and Developer Integration
Independent benchmarks published by Artifial Analysis place Muse Spark 1.2 on par with Grok 4.5, positioning both models behind Opus 5, Fable 5, Opus 4.8, GPT-5.6 Sol, and Kimi K3. Meta stated that version 1.2 delivers a moderate improvement over its predecessor, focusing optimization on developer agent tasks such as multi-file refactoring, extended debugging sessions, and multi-task management rather than raw benchmark scores. Developers can install Muse Code via a command-line interface on macOS and Linux systems, and the underlying Spark 1.2 model is also accessible through the OpenRouter platform.
>