[Recording Available!] Recorded live on May 27, 2026, San Francisco - Agents & APIs SF Developer Meetup. Demos by @public, @Firebase, @interfaze_ai & @getpostman. Check it out! https://t.co/sv1XRMhQNX
We've added the latest flash model to the suite of 9 benchmarks for tasks like OCR, Object detection, ASR
Interfaze outperforms Gemini 3.5 flash while being 2.6x cheaper
We don't really think Gemini 3.5 flash is truly a flash series model based on pricing but we still ran benchmarks
Interfaze outperforms on OCR & Object detection while being 3x less the cost
3.5 flash does better than 3 flash, so would any Pro model at that price range
Not sure why is Gemini flash 3.5 even considered a flash model based on its pricing
Seems pretty far away from what users would pay for production workflow use cases
And if the focus is coding, it seems far behind compared to gpt 5.5/opus 4.7 even with all that benchmaxing
Can a new AI architecture completely stop hallucinations? Here is a breakdown of how Interfaze works and how it performed against generalist models when parsing messy, real-world data.
🛑[LIVE] on day 3 of @aiDotEngineer 🇸🇬
@khurdula, co-founder and CTO of @interfaze_ai on building alternative architecture designed for deterministic developer tasks.
“The solution is: Task specific DNNs + CNNs + Transformers”
Interfaze is a new model architecture that outperforms in tasks like OCR, Object detection, Translation and more
It beats models like Claude Sonnet 4.6, Gemini 3 flash and GPT 5.4 mini on 9 benchmarks
Orchestrator implies a one way delegation or tool call but how it actually works is specialized DNN encoder layers which are Small models (SM) not SLM since they aren't language/transformer based layers.
The CNN/DNNs are encoded within the same vector space as the base transformer which act as the orchestrator kinda similar to MoE models but done with DNNs.
@bygregorr@aaron_epstein Both, benchmarks are industry standard way to have broad tasks but we work with customers with much worse data quality for extraction. We'll be dropping way me explains on different use cases! Stay tuned.