SquadStack launches sales-focused LLM, claims 96.7% accuracy on real sales calls

SquadStack has launched Maya, a large language model (LLM) designed for sales conversations, claiming it outperformed 10 models from OpenAI, Google, DeepSeek, Alibaba and other developers on its benchmark of real Indian sales calls.

The company said Maya achieved a 96.7% pass rate across 9,962 conversational exchanges drawn from 471 sales calls. The benchmark evaluated models on seven parameters, including their ability to maintain conversational context, follow campaign objectives and use tools correctly. Google’s Gemma 4 31B scored 92.5%, DeepSeek V4.1 Flash 92.2%, while OpenAI’s GPT-6 Luna scored 89%.

Maya also recorded a median response time of 278 milliseconds on SquadStack’s production endpoint in India. The company said the model was trained on more than 40 crore sales interactions collected over six years, with conversations tagged according to outcomes such as completed sales, callbacks and drop-offs.

According to SquadStack, Maya currently processes more than 50 lakh conversations daily in Hinglish and other Indian languages, handling over one trillion tokens a month.

“A live sales call is the hardest place to run an LLM,” said Vikas Gulati, Co-founder and CTO of SquadStack. “Thousands of calls can start in the same second, and the customer hears every slow turn. So we built Maya for this one job, with the sales behaviour in its weights rather than in a long prompt. Maya has already processed more than 5 trillion tokens, and it now handles over a trillion tokens a month.”

Maya is the company’s second model, following its speech recognition model Arth V2. It is deployed within SquadStack’s Voice AI agents for enterprise customers.

The company said the model is hosted in India and that its learning system, Lift, uses live call outcomes to identify and improve areas of performance.

AILLMMayaSalesSquadstack
Comments (0)
Add Comment