Sarvam-105B-1711 without Indic
Twelve Flash-Lite models on Sarvam’s chat and tool-calling benchmarks, next to price.
Score73.2#2 of 12
Blend 3:1Three input tokens for every output token, the usual chat mix.(3 × ₹10 in + ₹25 out) ÷ 4 = ₹13.75US$1 = ₹90, Sarvam’s own rate
Per 1M tokens₹13.75$0.15Score vs price
Average rank
Every benchmark
| Model | Avg | IFB | MC | AIF | WB | AH | τ² | BFCL |
|---|---|---|---|---|---|---|---|---|
| 76.0 | 80 | 65 | 62 | 87 | 81 | 86 | 72 | |
| 73.2 | 82 | 63 | 60 | 87 | 75 | 84 | 63 | |
| 73.1 | 74 | 65 | 65 | 81 | 77 | 83 | 68 | |
| 72.9 | 75 | 62 | 59 | 84 | 78 | 82 | 70 | |
| 72.5 | 76 | 62 | 59 | 84 | 75 | 80 | 72 | |
| 72.2 | 78 | 64 | 60 | 79 | 75 | 81 | 69 | |
| 70.1 | 58 | 62 | 60 | 85 | 74 | 82 | 70 | |
| 67.2 | 75 | 59 | 56 | 77 | 69 | 72 | 63 | |
| 65.4 | 71 | 51 | 42 | 87 | 74 | 78 | 55 | |
| 64.2 | 57 | 57 | 58 | 78 | 62 | 78 | 60 | |
| 61.0 | 66 | 47 | 40 | 84 | 66 | 71 | 53 | |
| 50.5 | 41 | 31 | 31 | 86 | 55 | 55 | 55 |
- IFBIFBench
- MCMultiChallenge
- AIFAdvanced-IF
- WBWritingBench
- AHArena-Hard v2
- τ²τ²-Bench
- BFCLBFCL v4
- WBWritingBench counts in Avg only
My take
- A very good Flash-Lite release, even with Indian languages set aside.
- It holds up against its rivals, and it’s a Flash-Lite model served in India that I can pay for in rupees.
- Excited for their bigger model, and glad it will be served here too.