Sunsetting Arbitrary Models: A Necessary Evolution for Graze Social
Overview
Graze Social is temporarily sunsetting our "textarbitrary" and "imagearbitrary" zero-shot classification features on October 5, 2025. We first announced this deprecation on September 14th to give feed builders time to migrate. While these tools have been valuable, they've become unsustainable for our infrastructure. We're replacing them with optimized default categories and will introduce an improved arbitrary classification system in the coming weeks.
Why This Change Is Necessary
Over the past few months, we've achieved significant infrastructure improvements:
- 60% reduction in average feed latency (time from post creation to feed inclusion)
- 70% reduction in operational costs
However, the arbitrary models have become a bottleneck:
- While only 2-5% of feeds use these features
- They consume 95% of our compute budget
- Each zero-shot classification requires hundreds of individual calculations per post, compared to about a dozen for our default models
This imbalance threatens the quality of service for our entire ecosystem, making this sunset necessary for Graze's long-term health.
Migration Path
Finding affected feeds
Simply log into to Graze Social - if your feeds are affected, you'll see an alert on your dashboard:
Clicking on that alert will take you to an account page that will directly list the affected feeds - go through them one by one to address the fixes needed. Either delete the rules, or make your categorizations exactly the text shown below in the new categories. Once completed, this dialog will disappear.
Alternative Filters
Text Similarity Node: Our recommended replacement
- Provide "anchor sentences" to test against post text and image alt text
- Achieves similar targeting capabilities safely and at scale
- Works for any text-based classification you can imagine
New Default Categories: Pre-trained classifications based on the most common use cases from arbitrary nodes
Available Categories
When the sunset occurs, feeds using arbitrary nodes will automatically evaluate against these new categories (assuming a score of 0 for non-matching content):
Text Categories:
[ "Academic & Intellectual", "Adult & Sexual Content", "AI & Machine Learning", "Animals", "Arts & Creative", "Aviation & Maritime", "Data & Computing", "Entertainment & Culture", "Film & TV", "Food & Beverages", "Food & Lifestyle", "Game Development", "Gaming", "Healthcare & Medicine", "Medical Education", "Medical Specialties", "Music", "Nature & Outdoors", "News & Media", "Politics", "Programming", "Society & Culture", "Software Development", "Sports", "Visual Arts" ]
Image Categories:
[ "Academic & Intellectual", "Adult & Sexual Content", "AI & Machine Learning", "Animals", "Arts & Creative", "Aviation & Maritime", "Data & Computing", "Entertainment & Culture", "Film & TV", "Food & Beverages", "Food & Lifestyle", "Game Development", "Gaming", "Healthcare & Medicine", "Medical Education", "Medical Specialties", "Music", "Nature & Outdoors", "News & Media", "Politics", "Programming", "Society & Culture", "Software Development", "Sports", "Visual Arts" ]
Timeline
- September 14, 2025: Deprecation notice first shown to users
- Now through October 5, 2025: Migrate your feeds to use text similarity nodes or default categories
- October 5, 2025: Arbitrary models will be turned off
- Coming weeks: New and improved arbitrary classification system will be launched
Our Commitment
We remain committed to providing best-in-class tools for feed builders on ATProto. This temporary sunset allows us to rebuild these features in a way that's sustainable and scalable for everyone in the Graze ecosystem. The new system will provide similar capabilities while maintaining the performance improvements we've achieved.
If you need assistance migrating your feeds, please reach out to our support team.