SquadStack Launches Maya, a Sales LLM Built and Hosted in India
SquadStack said Maya was post-trained on more than 40 crore real sales interactions and scored 96.7% on its own benchmark.
Key Takeaways
- SquadStack launched Maya, which it calls a Sales LLM, on November 16, 2026.
- The company said Maya was post-trained on more than 40 crore real sales interactions from six years.
- On SquadStack's own benchmark, Maya scored 96.7%, against 92.5% for Gemma 4 31B.
- SquadStack said Maya handles more than 50 lakh conversations a day in Hinglish and Indian languages.
- The company said Maya is built and hosted in India.
Table of Contents 7 sections
Noida-based SquadStack launched Maya on November 16, 2026. The company calls Maya a Sales LLM, a large language model built only for live sales calls. SquadStack said it post-trained Maya from an open-weight foundation model on more than 40 crore real sales interactions from six years of running sales calls.
SquadStack said each of those interactions is tagged with how the call ended: a sale, a callback or a drop-off. The company said Maya scored 96.7% on its own Telesales LLM Benchmark, ahead of Google's Gemma 4 31B at 92.5%. SquadStack also said Maya already handles more than 50 lakh conversations a day in Hinglish and Indian languages.
SquadStack's Maya launch at a glance
| Detail | Value |
|---|---|
| Company | SquadStack.ai, based in Noida |
| Product | Maya, which the company calls a Sales LLM |
| Launch date | November 16, 2026 |
| Training data | More than 40 crore real sales interactions from six years, as per the company |
| Score on SquadStack's own benchmark | 96.7% |
| Next best model on that benchmark | Gemma 4 31B at 92.5% |
| Daily conversations | More than 50 lakh, as per the company |
| Languages | Hinglish and Indian languages |
| Where it is built and hosted | India |
How SquadStack tested Maya on its own benchmark
The benchmark was built by SquadStack itself, and the scores are the company's own. SquadStack said it took moments from 471 real sales calls and asked eleven models to write the next reply an agent should give. Anthropic's Claude Haiku 5.5 was used as the judge. It marked each reply as a pass or a fail on seven checks, including staying on the goal of the call and using tools correctly.
As per SquadStack, Maya passed 96.7% of the checks. Gemma 4 31B scored 92.5%, DeepSeek V4.1 Flash scored 92.2% and OpenAI's GPT-6 Luna scored 89%. The company said Maya kept track of the conversation on 96% of turns, against 90% for the next best model, and picked the right reply 95% of the time, against 87%.
SquadStack also gave speed figures. It said Maya starts its reply in a median of 278 milliseconds on its own setup in India, and 95% of its replies start within 774 milliseconds. For Gemma 4 31B, the company said 95% of replies started within about 2.3 seconds.
Maya is SquadStack's second in-house model
SquadStack builds voice AI agents that make sales calls for businesses. Maya is the second model from the company's in-house model work, after its speech recognition model, Arth V2. The company said Maya runs inside its Voice AI Agents for enterprise customers, and new brands can request beta access.
The company said Maya is built and hosted in India, so customer conversations stay within the country.
Vikas Gulati on building a model for live sales calls
"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. So we built Maya for this one job, with the sales behaviour in its weights rather than in a long prompt."
What SquadStack said it is working on for Maya
SquadStack said its learning system, Lift, watches live calls and fixes what went wrong, and another system, Cerebro, carries each fix to other campaigns. The company said Maya will next read cues beyond words, such as a long pause, a change in tone or what happened on a customer's last call.
Where SquadStack's Maya stands at launch
Maya launched on November 16, 2026 with a 96.7% score on SquadStack's own Telesales LLM Benchmark, ahead of Gemma 4 31B at 92.5%. The company said the model was post-trained on more than 40 crore sales interactions and already handles more than 50 lakh conversations a day. SquadStack said Maya is built and hosted in India and is open to new brands through beta access.
More news
Frequently Asked Questions
What is SquadStack's Maya?
When did SquadStack launch Maya?
How was Maya trained?
How does SquadStack tag its sales interactions?
How did Maya score on SquadStack's benchmark?
Who built the benchmark used to test Maya?
How did other models do on SquadStack's benchmark?
How many conversations does Maya handle?
How fast does Maya reply?
Where is Maya hosted?
Who are the co-founders quoted on SquadStack's Maya launch?
Can businesses use SquadStack's Maya?
More in AI
Fractal Analytics Is in Buyout Talks With Up to 25 AI Startups
Fractal Analytics is in talks with 20 to 25 AI startups about possible acquisitions, its CEO said.
Amazon Launches Alexa for Shopping AI Assistant in India
Amazon has started rolling out Alexa for Shopping, a free AI shopping assistant, to customers in India.
Shield AI Opens an Engineering Office in Bengaluru
Shield AI has opened a Bengaluru engineering office to build Hivemind autonomy software for India.
