ACE & Agentic LLM Overview

Introduction to ACE LLM

ACE LLM by Gupshup is a family of generative models to automate marketing, commerce & support related conversations powered by Open & Closed source foundational models. It has been fine-tuned on industry & function-specific data to deliver a delightful customer experience. Enterprises get a head-start in delivering ACE LLM-powered conversational experiences.

The ACE LLM has been built keeping the focus on compliance, accuracy & cost efficiency. AI guardrails implemented in the ACE family ensure ethical use, safety in responses, preventing harmful outputs or biased information from delivering the accurate results.

Recently, We have upgraded our LLM models in to continuously improve the performance and accuracy of our AI systems.

ACE Family Models (Till 11.0 release (6 Oct 24))	ACE Family Models (12.0 -19.0 Release Starting from 7 Oct 24 - 10 Aug 25)	ACE Family Models (20.0 Release Starting from 11 Aug 25)
ACE LLM Lite (Flan t5, Llama 2)	ACE Lite: Llama-3.1-8b	ACE Lite: Qwen-3-4b
ACE LLM Hybrid (Flan T5, Azure Open AI)	ACE Pro: (Gemma-2-2b, Azure OpenAI GPT-4o-mini)	ACE Pro: (Gemma-2-2b, Azure OpenAI GPT-4o-mini)
ACE LLM Hybrid v2 (Mistral, Azure Open AI)	ACE Advanced: (Llama-3.1-8b, Azure OpenAI GPT-4o-mini)	ACE Advanced: (Qwen-3-4b, Azure OpenAI GPT-4o-mini)
ACE LLM Classic (Azure Open AI)	ACE Premium: Azure OpenAI GPT-4o-mini	ACE Premium: (Azure OpenAI GPT-4o-mini)

Use Case Handling

ACE Family	Imp. Use Case Served
ACE Lite	Qwen-3-4b: Intent & Entity Classification, FAQ Answering
ACE Pro	Gemma-2-2b: Intent & Entity Classification GPT-4o-mini: FAQ Answering
ACE Advanced	Qwen-3-4b: Intent & Entity Classification GPT-4o-mini: FAQ Answering
ACE Premium	GPT-4o-mini: Intent & Entity Classification, FAQ Answering

Introduction to Agentic LLM

Gemini & GPT models are the core engines that drive the intelligence, response generation, and decision-making capabilities of your AI agents within the Gupshup AI Admin Module.

These models are used during both intent classification & skill execution and are configurable at both agent and skill level.

Model	Description	Best For
GPT-4o (AI Agent Premium)	High-performance large language model from with strong reasoning & multi-modal capabilities	Complex use cases, longer replies, premium accuracy
GPT-4o Mini (AI Agent Pro)	Lightweight variant offering faster responses at a lower compute cost	Utility tasks, real-time replies, high-volume scenarios
Gemini 2.0 Flash (ACE Flash Agent Premium)	A powerful and versatile model with native multimodal input and a large 1M token context window for seamless, coherent conversations.	Complex conversational flows, in-depth chat applications, and multi-turn interactions.
Gemini 2.0 Flash-Lite (ACE Flash Agent Pro)	An optimized variant of Flash, designed for maximum speed and efficiency to deliver real-time, responsive conversational experiences at scale.	High-volume conversational tasks, real-time chatbots, and streamlined assistant applications.