Photo: Tima Miroshnichenko / Pexels
The company's new contributor tier slashes API costs by up to 21x if developers let Meta train on their usage data
Most AI companies let you opt out of sharing your data with a toggle and a privacy policy. Meta looked at that convention and decided to turn it into a business model.
The company launched Muse Spark 1.3 on September 2, offering two distinct pricing tiers for its latest coding-focused AI model. The standard tier runs $1.25 per million input tokens and $4.25 per million output tokens. The “Contributor” tier, meanwhile, drops those figures to $0.10 and $0.20 respectively. The catch: your prompts, responses, and usage patterns get fed back into Meta’s training pipeline.
The math behind the discount #
Those Contributor tier numbers represent a 12x to 21x discount depending on the token type. Cached input tokens, which let developers reuse context without reprocessing, fall from $0.15 per million on the standard plan to $0.002 on the Contributor plan. That’s a 75x reduction for cached inputs, making it almost trivially cheap to run repeated workflows.
The Contributor tier caps developers at 60 requests per minute, compared to 3,000 on the standard plan. That 50x reduction in throughput makes the discount tier essentially unusable for production-scale applications.
Why Meta needs your code #
Meta Superintelligence Labs, the division behind the Muse Spark series, built these models with a specific gap in mind. Meta’s core platforms, Instagram, Facebook, WhatsApp, generate enormous volumes of conversational and visual data. What they don’t generate is high-quality coding interactions.
Muse Spark 1.3 supports a context window of 1 million tokens, which means it can process roughly 750,000 words of context in a single session. The model also claims improved efficiency in reducing the number of tool calls needed to complete complex tasks compared to earlier versions.
The competitive landscape #
This pricing structure was first tested with Muse Spark 1.2 in August 2026, making the Contributor tier a deliberate strategic choice rather than a promotional experiment.
The strategy takes direct aim at OpenAI and Anthropic, both of which charge premium rates for their frontier models without offering comparable data-sharing discounts. OpenAI’s approach has been to collect training signal through its free ChatGPT tier, while Anthropic has generally maintained stricter data handling policies.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our