---
title: "NPCI models: release history and timeline"
description: "Not-for-profit company that operates India's retail payment rails, including UPI, and builds domain models for payments and banking."
type: "release-archive"
canonical_url: "https://theaiengineering.com/releases/npci"
markdown_url: "https://theaiengineering.com/releases/npci.md"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
updated: "2026-09-09"
citation: "The AI Engineering Editors. \"NPCI models: release history and timeline.\" The AI Engineering, updated Sep 09, 2026. https://theaiengineering.com/releases/npci"
license: "CC BY 4.0"
license_url: "https://creativecommons.org/licenses/by/4.0/"
data_csv: "https://theaiengineering.com/releases.csv"
data_json: "https://theaiengineering.com/releases.json"
---

# NPCI models: release history and timeline

Not-for-profit company that operates India's retail payment rails, including UPI, and builds domain models for payments and banking.

- Headquarters: Mumbai, India
- Website: https://www.npci.org.in

2 releases from 1 lab, 0 with open weights. Dates are the labs' official announcement dates.

## September 2026

### FiMI Banking

FiMI Banking is a compact NPCI model for Indian retail banking agents, built by post-training Google's Gemma 4 E4B to reason, call banking tools and complete multi-step tasks within a bank's rules.

- Date: 2026-09-09
- Lab: NPCI
- Type: Small model
- Weights: Proprietary
- Modalities: text → text
- Parameters: 4.5B effective
- Permalink: https://theaiengineering.com/releases/2026/09#npci-fimi-banking

Highlights:

- NPCI described it as India's first sovereign compact model for retail banking and said it would co-develop the model further with HDFC Bank
- The technical report says NPCI applied preference optimization and reinforcement learning with verifiable rewards to Gemma 4 E4B, which it lists at 4.5B effective parameters
- NPCI open-sourced two evaluation suites with it: IndicBank Bench (799 scripted multi-turn cases) and a Tau² agentic banking benchmark (1,000 tasks), both on GitHub under MIT
- Weights were not published at launch; the model targets deployment inside bank-controlled infrastructure, including air-gapped setups

Source: [NPCI unveils FiMI Banking, India's first sovereign compact model for Indian retail banking at GFF 2026; To Collaborate with HDFC Bank to Strengthen Tech](https://www.npci.org.in/uploads/NPCI_Press_Release_NPCI_unveils_Fi_MI_Banking_India_s_first_sovereign_compact_model_for_Indian_retail_banking_GFF_2026_To_Collaborate_with_HDFC_Bank_to_Strengthen_Tech_c0642e7bbc.pdf)

## February 2026

### FiMI

FiMI (Finance Model for India) is NPCI's domain model for India's payments system, adapted from Mistral Small 24B to handle UPI transactions, disputes, mandates and regulatory questions.

- Date: 2026-02-17
- Lab: NPCI
- Type: LLM
- Weights: Proprietary
- Modalities: text → text
- Parameters: 24B
- Available at launch: Consumer app
- Permalink: https://theaiengineering.com/releases/2026/02#npci-fimi

Highlights:

- NPCI continued pretraining Mistral Small 24B Base on 68 billion tokens of financial, multilingual and synthetic data, then fine-tuned FiMI Base and FiMI Instruct variants
- NPCI reported a 20% gain over Mistral Small 24B Base on finance reasoning and an 87% gain over Mistral Small 24B Instruct on domain tool calling
- At launch it powered NPCI's UPI Help Assistant, which answered UPI user queries in English, Hindi, Telugu and Bengali
- NPCI published the training method and evaluation on arXiv but did not release weights

Source: [NPCI announces FiMI, a domain-specific AI language Model built for India's payments ecosystem](https://www.npci.org.in/uploads/NPCI_announces_Fi_MI_a_domain_specific_AI_language_Model_built_for_India_s_payments_ecosystem_fce0d71c70.pdf)

## Cite and reuse this data

These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL. Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".

The AI Engineering Editors. "NPCI models: release history and timeline." The AI Engineering, updated Sep 09, 2026. https://theaiengineering.com/releases/npci

HTML link:

```html
<a href="https://theaiengineering.com/releases/npci">NPCI models: release history and timeline</a> (The AI Engineering)
```

- Title: NPCI models: release history and timeline
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- Updated: 2026-09-09
- URL: https://theaiengineering.com/releases/npci
- Markdown: https://theaiengineering.com/releases/npci.md
- License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
- Full dataset as CSV: https://theaiengineering.com/releases.csv
- Full dataset as JSON: https://theaiengineering.com/releases.json
