Sarvam AIproprietary
Released: 2025-06-15Sarvam 105B
Sarvam AI flagship 105B frontier model optimized for complex reasoning, multi-turn dialogue, and cultural nuances across all 22 official Indian languages and English.
Intelligence Index
88.5/ 100
Calibrated multi-domain compositeThroughput Speed
72tok/sec
Output streaming throughputTime To First Token
340ms
Initial chunk server latencyLiveBench Contam-Free
78.4%
September 2026 suite scoreSourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 56.7% | 93 | 2025-07-20 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 68% | 75 | 2025-07-13 | independent | Source ↗ |
| FrontierMath | Active | 23.7% | 98 | 2025-07-10 | independent | Source ↗ |
| PutnamBench | Active | 43.9% | 90 | 2025-07-09 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 55.1% | 20 | 2025-07-07 | independent | Source ↗ |
| Omni-MATH | Deprecated | 62.5% | 10 | 2025-07-07 | independent | Source ↗ |
| Soohak | Active | 64.2% | 98 | 2025-07-05 | independent | Source ↗ |
| FrontierCode | Active | 47% | 98 | 2025-07-05 | independent | Source ↗ |
| Humanity's Last Exam | Active | 38.4% | 93 | 2025-07-03 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 69.2% | 75 | 2025-07-03 | independent | Source ↗ |
| AIME | Nearing Saturation | 80.3% | 75 | 2025-07-03 | independent | Source ↗ |
| OlympiadBench | Active | 69.3% | 80 | 2025-07-02 | independent | Source ↗ |
| MathArena | Active | 67.6% | 99 | 2025-07-01 | independent | Source ↗ |
| FireBench | Active | 68.4% | 95 | 2025-07-01 | independent | Source ↗ |
| Terminal-Bench | Active | 56.7% | 80 | 2025-07-01 | independent | Source ↗ |
| OpenAI-MRCR | Active | 74.3% | 80 | 2025-06-30 | independent | Source ↗ |
| SimpleQA | Active | 46% | 90 | 2025-06-30 | independent | Source ↗ |
| CodeContests | Deprecated | 52.7% | 10 | 2025-06-30 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 66.8% | 80 | 2025-06-29 | independent | Source ↗ |
| SWE-bench Pro | Active | 50.2% | 90 | 2025-06-29 | independent | Source ↗ |
| LongBench-Pro | Active | 72.6% | 80 | 2025-06-29 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 70.8% | 78 | 2025-06-28 | independent | Source ↗ |
| MultiPL-E | Deprecated | 74.5% | 10 | 2025-06-28 | independent | Source ↗ |
| LiveCodeBench | Saturated | 66.4% | 39 | 2025-06-27 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 55.4% | 80 | 2025-06-27 | independent | Source ↗ |
| MATH | Saturated | 82.8% | 25 | 2025-06-27 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 75.2% | 65 | 2025-06-27 | independent | Source ↗ |
| Mercury | Deprecated | 71.3% | 10 | 2025-06-26 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 70.8% | 65 | 2025-06-26 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 81.4% | 75 | 2025-06-26 | independent | Source ↗ |
| RULER | Nearing Saturation | 81.4% | 84 | 2025-06-25 | independent | Source ↗ |
| TabMWP | Saturated | 77.7% | 25 | 2025-06-25 | independent | Source ↗ |
| CyberSecEval | Deprecated | 72.9% | 10 | 2025-06-25 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 63.2% | 25 | 2025-06-24 | independent | Source ↗ |
| MMLU-Redux | Saturated | 86.7% | 25 | 2025-06-24 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 77.9% | 78 | 2025-06-24 | independent | Source ↗ |
| LongBench | Nearing Saturation | 76.1% | 65 | 2025-06-23 | independent | Source ↗ |
| SWE-bench | Deprecated | 55.1% | 10 | 2025-06-23 | independent | Source ↗ |
| MGSM | Nearing Saturation | 80.3% | 65 | 2025-06-23 | independent | Source ↗ |
| AGIEval | Saturated | 75% | 25 | 2025-06-22 | independent | Source ↗ |
| QASPER | Nearing Saturation | 73.5% | 65 | 2025-06-22 | independent | Source ↗ |
| TruthfulQA | Saturated | 77.9% | 25 | 2025-06-22 | independent | Source ↗ |
| GSM-Hard | Deprecated | 81.1% | 10 | 2025-06-21 | independent | Source ↗ |
| SuperGLUE | Saturated | 88.9% | 25 | 2025-06-21 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 76.1% | 23 | 2025-06-20 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 8.4% | 75 | 2025-06-20 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 85.5% | 25 | 2025-06-20 | independent | Source ↗ |
| WinoGrande | Saturated | 90.7% | 28 | 2025-06-19 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 82.6% | 65 | 2025-06-19 | independent | Source ↗ |
| IFEval | Deprecated | 79.8% | 25 | 2025-06-19 | independent | Source ↗ |
| GPQA | Nearing Saturation | 77.9% | 78 | 2025-06-19 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 97.7% | 44 | 2025-06-18 | independent | Source ↗ |
| HumanEval+ | Saturated | 87.5% | 25 | 2025-06-18 | independent | Source ↗ |
| CommonsenseQA | Saturated | 91.6% | 28 | 2025-06-18 | independent | Source ↗ |
| MBPP | Saturated | 85.9% | 25 | 2025-06-17 | independent | Source ↗ |
| GSM8K | Saturated | 91.3% | 25 | 2025-06-17 | independent | Source ↗ |
| GLUE | Saturated | 94.2% | 28 | 2025-06-17 | independent | Source ↗ |
| MMLU | Saturated | 90.3% | 25 | 2025-06-16 | independent | Source ↗ |
| HumanEval | Saturated | 92.3% | 25 | 2025-06-16 | independent | Source ↗ |
| ARC-AGI | Saturated | 82% | 25 | 2025-06-16 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 78.4% | 94 | 2025-02-01 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="sarvam-105b",
messages=[{"role": "user", "content": "Evaluate multi-step logic problem"}],
temperature=0.2
)
print(response.choices[0].message.content)Evaluation & Sourcing Notes
Scores listed for Sarvam 105B represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.