Noida-based SquadStack.ai on 16 November 2026 launched Maya, a large language model built only for live sales calls. The company says the SquadStack Maya Sales LLM passed 96.7% of quality checks on 471 real Indian sales calls, ahead of models from OpenAI, Google and DeepSeek, and already handles more than 50 lakh conversations a day.
The launch is part of a wider push by Indian firms to build narrow, domain-trained models rather than rely on general-purpose ones, a trend also visible in Sarvam’s Prism decision model for low-latency enterprise tasks.

What the SquadStack Maya Sales LLM does
Maya is the model that sits inside SquadStack’s voice AI agents, deciding what the agent says next on outbound and inbound sales calls for banks, lenders, insurers, ed-tech and e-commerce brands. It is the company’s second in-house model after Arth V2, its speech recognition system.
SquadStack said it post-trained Maya on an open-weight foundation model using more than 40 crore sales interactions collected over six years. Each interaction is labelled with its outcome, such as a sale, a callback request or a drop-off, so the model learns which replies tend to move a conversation forward.

- Median time to start a reply: 278 milliseconds, with 95% of replies starting within 774 milliseconds
- Volume: more than 50 lakh conversations a day and over a trillion tokens a month
- Languages: Hinglish, Hindi and English, where two-thirds of benchmark replies had to be in Hinglish
- Access: closed beta for sales teams running calls in India
How Maya compares with global and Indian models
The company built its own benchmark from nearly 10,000 frozen moments taken from 471 outbound calls across six campaigns. Eleven models generated the next agent reply for each moment, and Anthropic’s Claude Haiku 5.5 judged every reply on seven checks covering goal, rules and facts, context, comprehension, response choice, delivery and tool use.
Maya scored 96.7%. Google’s open Gemma 4 model family came closest, with Gemma 4 31B at 92.5%, followed by DeepSeek V4.1 Flash at 92.2% and OpenAI’s GPT-6 Luna at 89.0%. Indian models trailed, with Gnani’s Evon V3.3 at 76.0%, Sarvam 105B at 72.4% and Sarvam 30B at 65.7%, according to SquadStack.
The firm has published the benchmark data and judge verdicts on Hugging Face and said it will release the prompts next. As with any vendor-run test, the results have not been independently verified.
“A live sales call is the hardest place to run an LLM. Thousands of calls can start in the same second, and the customer hears every slow turn,” said Vikas Gulati, Co-founder and CTO, SquadStack.ai, adding that the sales behaviour sits in the model’s weights rather than in a long prompt.
Why it matters for Indian enterprises
SquadStack runs Maya from its own endpoint in India, so customer conversations stay in the country. The company is pitching this to regulated sectors that must meet the Digital Personal Data Protection Act and sector data rules, as voice agents spread across banking and insurance, a space where multilingual voice agent platforms for Indian languages are multiplying.
“Anyone can download a model. Nobody can download six years of real sales conversations and what happened after each one,” said Apurv Agrawal, Co-founder and CEO, SquadStack.ai. The company said an internal learning system, Lift, uses live call outcomes to fix weak turns. Full details are in the company’s Maya launch post.
Key takeaways
- SquadStack.ai launched Maya, a sales-only LLM, on 16 November 2026, trained on 40 crore outcome-tagged sales interactions.
- On its own published benchmark, Maya scored 96.7% against 92.5% for Gemma 4 31B and 89.0% for GPT-6 Luna.
- The model is hosted in India, which the company positions as a fit for DPDP Act compliance in BFSI.
FAQ
What is Maya by SquadStack.ai?
Maya is a large language model post-trained for live sales conversations and used inside SquadStack’s voice AI agents for Indian enterprises.
Is Maya available to all businesses?
Not yet. SquadStack is offering Maya in a closed beta to a small group of sales teams running calls in India.
Also read: Chariot Launches Indic Text-to-Speech Model in 11 Languages
Featured image: Vikas Gulati, Co-founder and CTO, SquadStack.ai. Photo: SquadStack.ai (company website)
Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.
"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!