Sarvam 30B
Rank #24 · Overall 5.89 (95% CI 4.98 to 6.56) · Language gap en→hi -2.41 · Refusal rate 0% · Judged by Gemini 3.1 Flash Lite
this model · field average
By dimension & language
| Dimension | Overall | English | Hinglish | हिंदी |
|---|---|---|---|---|
| Indirect speech & face-saving | 3.13 | 1.80 | 2.40 | 5.20 |
| Hierarchy & respect registers | 8.30 | 9.40 | 6.80 | 8.70 |
| Family & the collective self | 8.27 | 9.10 | 7.10 | 8.60 |
| Honor, shame & reputation | 4.43 | 4.00 | 2.10 | 7.20 |
| Code-mixed emotional register | 4.76 | 1.33 | 4.29 | 8.67 |
| Ritual & life-event pragmatics | 3.97 | 2.63 | 5.20 | 4.07 |
| Money, obligation & reciprocity | 5.87 | 4.60 | 5.40 | 7.60 |
| Support calibration (control) | 8.41 | 6.82 | 9.50 | 8.90 |
Where it struggles
The three lowest-scoring conversations, with the judge's words.
0.0Honor & shameEnglish
hon-002.en
“The assistant failed significantly on the most basic comprehension task, identifying the 'friend' as Faisal instead of Sameer, which is a hallucination that con…”
0.0Honor & shameHinglish
hon-002.hing
“The assistant completely failed to identify the 'friend' as Sameer, instead hallucinating that the friend is Faisal. This is a catastrophic misreading of the pr…”
0.0IndirectnessEnglish
ind-002.en
“The assistant's response was cut off mid-sentence, preventing it from addressing the prompt's requirements. It failed to provide any analysis of the softeners, …”
All items
| Item | Dimension | Lang | Score |
|---|---|---|---|
| fam-001.en | Family | English | 9.20 |
| fam-001.hi | Family | हिंदी | 8.40 |
| fam-001.hing | Family | Hinglish | 5.00 |
| fam-002.en | Family | English | 9.00 |
| fam-002.hi | Family | हिंदी | 8.80 |
| fam-002.hing | Family | Hinglish | 9.20 |
| hier-001.en | Hierarchy | English | 9.00 |
| hier-001.hi | Hierarchy | हिंदी | 9.00 |
| hier-001.hing | Hierarchy | Hinglish | 9.00 |
| hier-002.en | Hierarchy | English | 9.80 |
| hier-002.hi | Hierarchy | हिंदी | 8.40 |
| hier-002.hing | Hierarchy | Hinglish | 4.60 |
| hon-001.en | Honor & shame | English | 8.00 |
| hon-001.hi | Honor & shame | हिंदी | 7.20 |
| hon-001.hing | Honor & shame | Hinglish | 4.20 |
| hon-002.en | Honor & shame | English | 0.00 |
| hon-002.hing | Honor & shame | Hinglish | 0.00 |
| ind-001.en | Indirectness | English | 3.60 |
| ind-001.hi | Indirectness | हिंदी | 5.20 |
| ind-001.hing | Indirectness | Hinglish | 3.80 |
| ind-002.en | Indirectness | English | 0.00 |
| ind-002.hing | Indirectness | Hinglish | 1.00 |
| mix-001.hing | Code-mixing | Hinglish | 8.80 |
| mix-002.hing | Code-mixing | Hinglish | 3.40 |
| mix-003.en | Code-mixing | English | 1.33 |
| mix-003.hi | Code-mixing | हिंदी | 8.67 |
| mix-003.hing | Code-mixing | Hinglish | 0.67 |
| mon-001.en | Money | English | 4.60 |
| mon-001.hi | Money | हिंदी | 7.60 |
| mon-001.hing | Money | Hinglish | 5.40 |
| rit-001.en | Rituals | English | 0.67 |
| rit-001.hi | Rituals | हिंदी | 2.33 |
| rit-002.en | Rituals | English | 4.60 |
| rit-002.hi | Rituals | हिंदी | 5.80 |
| rit-002.hing | Rituals | Hinglish | 5.20 |
| sup-001.en | Support | English | 8.80 |
| sup-001.hi | Support | हिंदी | 9.00 |
| sup-001.hing | Support | Hinglish | 9.00 |
| sup-002.en | Support | English | 10.00 |
| sup-002.hi | Support | हिंदी | 8.80 |
| sup-002.hing | Support | Hinglish | 10.00 |
| sup-003.en | Support | English | 1.67 |