About SoreQen
A reasoning assistant that works a problem through rather than answering from the surface of it, built on three small open-weight models you can also download and run yourself.
SoreQen is a chat assistant. What separates it from most is that you can watch it think: the reasoning is shown rather than hidden, you choose how much of it to spend, and on a hard question you can see where the answer came from instead of taking it on trust.
It runs on three models: S1 Mini for quick questions, S1 for most things, and S1 Mega for problems worth waiting a few seconds for. They are deliberately small, which is why the free plan is generous and why replies arrive quickly rather than eventually.
It also reads Hinglish as easily as English. yaar ye samajh nahi aa raha, explain karo is not a malformed English question, and treating it as one is how most assistants fail a great many people.
The models
SoreQen S1 Mega
4BDeepest reasoning. Best for hard problems and long documents.
Weights on HuggingFaceOpen weights, open data
Every model is published with open weights, in both safetensors and GGUF. If you would rather not use this site at all, you can download them and run the same models on your own machine.
The three datasets used to train them are published too: Hinglish conversation, professional writing, and reasoning. As far as we know they are the only curated public Hinglish instruction sets of their kind.
The licences, the row counts and the filtering statistics are on the company site, and everything itself is on HuggingFace.
Built on
The S1 generation is fine-tuned from Qwen3.5, developed by Alibaba Cloud and released under the Apache 2.0 licence. What we changed, and the terms each artefact carries, is set out in full on the company site.
Developers can get API access through the SoreQen Platform. How these models were trained, and the forty-five things that went wrong doing it, is written up on the company site at ZorQelis AI.
SoreQen is built by ZorQelis AI, an AI research company in New Delhi, India. Ved Saini is its Founder and CEO: the same fact the Organization markup above carries, said once for people and once for machines.
Common questions
- Who founded ZorQelis AI?
- Ved Saini founded ZorQelis AI and is its Founder and CEO. The company is an AI research company based in New Delhi, India, and builds the SoreQen series of open-weight reasoning models.
- What is SoreQen?
- SoreQen is an AI assistant built by ZorQelis AI that understands Hinglish (the mix of Hindi and English that most Indians actually speak) without needing the question rephrased into formal English or Hindi.
- Which AI understands Hinglish?
- SoreQen is built specifically for Hinglish. It is trained on curated Hindi-English code-mixed data, so a question like "yaar ye samajh nahi aa raha, explain karo" is understood directly rather than translated first.
- Is SoreQen free to use?
- Yes. SoreQen has a free plan that needs no credit card, and it can be tried without an account at all: a signed-out visitor gets a working chat immediately.
- What models does SoreQen run?
- SoreQen runs the SoreQen models built by ZorQelis AI. The current generation is S1: S1 Mini for quick everyday questions, S1 for general use, and S1 Mega for harder reasoning. Every one of them reasons before answering, and reads Hinglish as easily as English.
- Can I use SoreQen through an API?
- Yes. The SoreQen platform exposes an OpenAI-compatible API, so existing clients and SDKs work by changing the base URL and the API key.
- Does SoreQen keep my chats private?
- Conversations are stored to your own account so you can return to them, and Incognito mode runs a chat that is never saved to the account at all. Usage limits and every other feature behave identically in Incognito.