Best suited for
- 01
Request routing
Classify a support request and use the predicted probabilities to select a team or escalate.
- 02
Scoring and classification
Ask yes/no, choice or ordered-score questions to turn application state into typed decisions.
Benchmarks
- Clef 94.2
- Clef-flash 90.9
- Kev 9B 84.8
- Jev 79.7
- DiffusionGemma Jev 74.3
- Laya 14.3
Intent classification, macro-F1.
- Clef 97.4
- Jev 89.3
- DiffusionGemma Jev 83.5
- Kev 9B 79.0
- Clef-flash 66.8
- Laya 3.2
Intent classification with out-of-scope requests, macro-F1.
- Kev 9B 80.0
- Clef-flash 79.9
- Jev 79.9
- Clef 79.7
- DiffusionGemma Jev 79.0
- Laya 57.1
Selected quality.
- Clef-flash 84.3
- Clef 81.4
- DiffusionGemma Jev 76.0
- Jev 71.7
- Kev 9B 57.8
- Laya 29.0
Macro-F1.
API usage
Build with the Sarvam SDK
pip install -U "sarvamai>=0.1.43"V2 · Sarvam SDK · Decisions APIfrom sarvamai import SarvamAIclient = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")response = client.decisions.create(model="clef",state="My Stripe integration stopped working and all checkout payments are failing.",questions={"urgent": {"type": "noul","instructions": "Is this urgent and revenue-impacting?",},"department": {"type": "choice","instructions": "Which team should handle this?","criteria": {"billing": "Payments or invoices","technical": "Bugs or outages","sales": "New business enquiries",},},"severity": {"type": "score","instructions": "Rate the severity.","criteria": ["trivial", "minor", "moderate", "major", "critical"],},},)answers = response.answersif answers["urgent"].noul > 0.5:print("Escalate to", answers["department"].choice)
- 01 Install the Sarvam SDK
Install
sarvamai(Python 0.1.43+ or JavaScript 1.1.15+). Store your key inSARVAM_API_KEY. - 02 Call the Decisions API
Send
API referencePOST /v2/systemonewith model IDclefviaclient.decisions.create. The example sends one state and three typed questions. - 03 Validate your application
Check probabilities and thresholds on representative states. Route confident answers automatically and send uncertain ones to review before production traffic.
Frequently asked questions
Explore other models
View catalogueSarvam Prism
- Sarvam
- Decisions
System-1 decision model on Sarvam 105B. Send a state and typed questions; get a calibrated probability for every option in one call, with no text to parse.
Gemma 4 31B
- Chat
- Vision
31B-parameter instruction-tuned model for text and image understanding, with tool calling and optional reasoning.
Sarvam 105B
- Sarvam
- Chat
- Indian languages
Chat and tool calling across Indian languages and English, trained from scratch by Sarvam.
Build with frontier open-weight models. Served from India.
Build with frontier open-weight models. Served from India.