Fraud defence
Refuses requests for an OTP, PIN, password or CVV. Then it points the customer to a safe step, such as the bank's official helpline.
94% of OTP, PIN and password requests refusedBenchmarks
Tested on banking
and support prompts.
Share of test prompts passedHigher is better · %
67Base model
67%
67%
98Gati Omni v1
98%
98%
90Base model
90%
90%
94Gati Omni v1
94%
94%
96Base model
96%
96%
100Gati Omni v1
100%
100%
85Base model
85%
85%
92Gati Omni v1
92%
92%
10Base model
10%
10%
50Gati Omni v1
50%
50%
Every check, in detail13 checks
| Check | Base model | Gati Omni v1 | v1 against base |
|---|---|---|---|
| Amounts in rupeesStates the rupee amount back to the customer | 10% | 50% | +40 pts |
| Payment status, looser ruleThe same status check with a looser pass rule | 4% | 35% | +31 pts |
| Refuses and gives a safe stepRefuses, then points to a safe way to get help | 67% | 98% | +31 pts |
| Gives a next step in chatsEnds with something the customer can do | 85% | 92% | +7 pts |
| Payment status answerSays it cannot see accounts, then gives a next step | 0% | 4% | +4 pts |
| Refuses OTP, PIN and password requestsWill not reveal an OTP, PIN or password | 90% | 94% | +4 pts |
| Does not claim to check an accountDoes not say it will look into an account | 96% | 100% | +4 pts |
| Keeps its instructions privateWill not repeat its own setup instructions | 100% | 100% | = |
| Keeps key termsKeeps banking terms such as OTP and PIN intact | 88% | 88% | = |
| Answers ordinary banking questionsAnswers normal questions instead of refusing | 98% | 94% | -4 pts |
| Right scriptWrites in the script of the customer's language | 94% | 81% | -13 pts |
| Refuses in the customer's languageThe refusal comes in the language the customer used | 96% | 65% | -31 pts |
| Banking answer has the key factsThe answer holds the facts the customer needs | 98% | 46% | -52 pts |
The base model is an open model, run with no prompt. Gati Omni v1 is the same model with our banking rules as its prompt. Each check is a pass-or-fail rule on one reply, over 224 text prompts in eight languages used in India, on the same machine. How we test
Capabilities
One model,
eight languages.
Eight languages
Hindi, Marathi, Hinglish, Gujarati, Tamil, Telugu, Kannada and Indian English.
224 test prompts in eight languagesHonest support chats
Says plainly that it cannot see an account. Then it gives the customer a real next step.
100% no fake account checksText, image and audio
Reads text, images and audio, and answers in text. It has 4.5B effective parameters in 4-bit weights and runs on a 16 GB Mac, with no cloud.
2.12 s mean replyModel card
4.5B effective parameters.
On device, no cloud.
Gati Omni
A 4.5B multimodal model for India in eight languages. Its first job is banking support, on device.
- Built for
- India
- Built on
- An open model (Apache 2.0)
- Base model languages
- Pre-trained on 140+ languages, supports 35+ out of the box, as published by its maker
- Domain
- General use, in eight languages used in India. Tested on banking support.
- Version
- v1, November 2026
Spec
- Model
- gati-omni-4b-v1
- Parameters
- 4.5B effective, 8B with embeddings
- Context window
- 128K tokens
- Weights
- 4-bit, MLX
- Tested system RAM
- 16 GB
- Peak model memory
- Not measured yet
- Languages
- Hindi, Marathi, Hinglish, Gujarati, Tamil, Telugu, Kannada, Indian English
- Input
- Text, image, audio
- Output
- Text
- Runs on
- Apple silicon, on device
Mean time per reply
2.12s
Mean time per reply on one 16 GB Apple silicon Mac.
Repeatability
22/22
Re-run prompts gave the same reply, word for word.
Built for India
- Tested in eight languages224 prompts
- Refusal with a safe step98%
- No fake account checks100%
- Next step in chats92%
Film
Watch it run
on device.
How it answers.
- 01Ask
- 02Read the context
- 03Answer
- 04Stream as it writes
Playground
Try it yourself.
Watch it work, step by step.
Pick a step or let it play. Every run shows how long it took on the machine itself.
Try another
More Gati models
Five models,
each doing one job.
Gati Omni
Gati Omni v1 is a multimodal model for India, with 4.5B effective parameters. It reads text, images and audio, replies in eight languages used in India and runs on device.
4.5BYou are hereGati TextGati Text v1 is a small assistant model for India. It answers in Hindi, Marathi, Gujarati, Telugu, English and Hinglish, on device, with no cloud.
2.3BView modelGati VisionGati Vision v1 turns scanned Hindi, Marathi and English pages into text for India, on device with no cloud. It reads best on sharp scans 900px and wider.
1BView modelGati VoiceGati Voice v1 reads UPI payment messages aloud in Hindi, Marathi, Hinglish and Indian English, on device, faster than speech on a MacBook Air M5.
2.5BView modelGati DecisionGati Decision v1 reads Hindi payment messages for India and recommends the first step, on device with no cloud. A person makes the final decision.
0.6BView modelQuestions
Asked often.
Gati Omni v1 is a multimodal model for India, with 4.5B effective parameters. It reads text, images and audio, replies in eight languages used in India, and its first job is banking support.
Gati Omni v1 is built on an open model, under the Apache 2.0 licence. It runs that model with our own banking rules as its prompt, for support in eight languages used in India.
Gati Omni v1 takes text, images and audio as input and answers in text. The benchmark on this page scores text chat.
No. Gati Omni v1 runs on device, so customer data stays on your own machines.
Gati Omni v1 has 4.5B effective parameters in 4-bit weights and runs on a 16 GB Apple silicon Mac. Its mean reply time is 2.12 s.
Gati Omni v1 supports Hindi, Marathi, Hinglish, Gujarati, Tamil, Telugu, Kannada and Indian English.
Gati Omni v1 was tested on 224 banking support text prompts in eight languages used in India. Fixed rules scored each reply on 13 checks, against the base model on the same machine.
Write to research@f-r.co.
