Recommends the first step
Reads a Hindi payment message and recommends what to do first: wait, raise a dispute, report fraud or nothing.
90% on harder messagesBenchmarks
Built for India.
Better on hard cases.
Pass rate by checkHigher is better · % of messages right
63Base model
63%
63%
90Gati Decision v1
90%
90%
80Base model
80%
80%
93Gati Decision v1
93%
93%
90Base model
90%
90%
93Gati Decision v1
93%
93%
100Base model
100%
100%
100Gati Decision v1
100%
100%
93Base model
93%
93%
93Gati Decision v1
93%
93%
Every check, in detail6 checks
| Check | Base model | Gati Decision v1 | v1 against base |
|---|---|---|---|
| Harder messages, what to do firstPicks the first step on less obvious messages | 63% | 90% | +27 pts |
| Clear messages, what to do firstPicks the first step on plain, simple messages | 90% | 93% | +3 pts |
| UrgencyJudges how urgent the customer message is | 100% | 100% | = |
| Money stuck, sheet oneSpots when the customer's money is stuck | 93% | 100% | +7 pts |
| Money stuck, sheet twoSpots stuck money on a second set of messages | 93% | 86% | -7 pts |
| Failed but debitedSpots payments that failed but still took money | 80% | 93% | +13 pts |
Same messages and same question wording for the base model and Gati Decision v1. The chart shows every check, with both money stuck sets together. The table below lists each set on its own. Each check uses 14 to 30 Hindi messages. The last column is v1 against base. How we test
Capabilities
One model,
four jobs done well.
Spots stuck money
Spots when a customer's money is stuck in a payment.
100%, 15 of 15 rightRecommends, people decide
It scores your options and recommends one. A person on your team makes the final call.
0 words of free textRuns on device
A 0.6B model that trains and runs on a MacBook Air with 16 GB. No cloud.
About 21 min to trainModel card
0.6B parameters.
Runs on one laptop.
Gati Decision
Gati Decision v1 is a 0.6B model for India that reads a Hindi payment message, scores the next steps on device and recommends one for a person to approve.
- Built for
- India
- Weights
- Pinned by SHA-256 hash
- Domain
- Payment problems in India
- Version
- v1, November 2026
Spec
- Model
- gati-decision-0.6b-v1
- Parameters
- 0.6B
- Language
- Hindi
- Input
- A message and a list of options
- Output
- A probability per option
- Ran on
- MacBook Air M4, 16 GB
- Peak model memory
- Not measured yet
- Training
- About 21 min, one run
- Cloud
- Not needed
Harder messages, what to do first
90%
27 of 30 harder messages right, against 19 for the base model.
Clear messages, right first step
28/30
28 of 30 plain Hindi messages got the right first step.
Built for India
- Harder messages, what to do first90%
- Clear messages, what to do first93%
- Money stuck, sheet one100%
- Harder messages fixed against broken9 to 1
Film
Watch it run
on device.
How it decides.
- 01Read
- 02List the options
- 03Score
- 04Pick the top one
Playground
Try it yourself.
Watch it work, step by step.
Pick a step or let it play. Demos tagged as illustrations show example output, and the scores on this page come from our own tests.
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.
4.5BView modelGati 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.
0.6BYou are hereQuestions
Asked often.
Gati Decision v1 is a 0.6B model for banks and payment apps in India. It reads a Hindi customer message about a payment problem and recommends the first step: wait, raise a dispute, report fraud or nothing.
No. Gati Decision v1 runs on device. It was trained on one MacBook Air M4 with 16 GB of memory.
No. Gati Decision v1 scores your options and recommends one. A person on your team makes the final call, and the model writes no text of its own.
Gati Decision v1 reads Hindi customer messages. Gati models are built for India.
Gati Decision v1 was tested on Hindi customer messages about payment problems, against the base model on the same messages. On harder messages it picks the right first step 90% of the time, against 63% for the base model.
Write to research@f-r.co.
