Sarvam AIproprietary
Released: 2025-04-10

Sarvam 30B

High-throughput multilingual enterprise model delivering high reasoning fidelity and accurate transliteration across Indic languages at low latency.

Intelligence Index
84.2/ 100
Calibrated multi-domain composite
Throughput Speed
95tok/sec
Output streaming throughput
Time To First Token
260ms
Initial chunk server latency
LiveBench Contam-Free
74.0%
September 2026 suite score

Sourced Benchmark Evaluations (61)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
ARC-AGI-3Active53.8%
93
2025-05-15independentSource ↗
ARC-AGI-2Nearing Saturation64.6%
75
2025-05-08independentSource ↗
FrontierMathActive22.4%
98
2025-05-05independentSource ↗
PutnamBenchActive41.6%
90
2025-05-04independentSource ↗
SWE-LancerDeprecated52%
20
2025-05-02independentSource ↗
Omni-MATHDeprecated59.2%
10
2025-05-02independentSource ↗
SoohakActive60.8%
98
2025-04-30independentSource ↗
FrontierCodeActive44.4%
98
2025-04-30independentSource ↗
Humanity's Last ExamActive36.4%
93
2025-04-28independentSource ↗
StructFlowBenchNearing Saturation65.8%
75
2025-04-28independentSource ↗
AIMENearing Saturation76%
75
2025-04-28independentSource ↗
OlympiadBenchActive65.6%
80
2025-04-27independentSource ↗
MathArenaActive64%
99
2025-04-26independentSource ↗
FireBenchActive65.1%
95
2025-04-26independentSource ↗
Terminal-BenchActive53.6%
80
2025-04-26independentSource ↗
OpenAI-MRCRActive70.7%
80
2025-04-25independentSource ↗
SimpleQAActive43.8%
90
2025-04-25independentSource ↗
CodeContestsDeprecated49.7%
10
2025-04-25independentSource ↗
ComplexBenchNearing Saturation63.5%
80
2025-04-24independentSource ↗
SWE-bench ProActive47.4%
90
2025-04-24independentSource ↗
LongBench-ProActive69%
80
2025-04-24independentSource ↗
ZeroSCROLLSNearing Saturation67.4%
78
2025-04-23independentSource ↗
MultiPL-EDeprecated70.4%
10
2025-04-23independentSource ↗
LiveCodeBenchSaturated62.7%
39
2025-04-22independentSource ↗
AgentIFNearing Saturation52.7%
80
2025-04-22independentSource ↗
MATHSaturated78.4%
25
2025-04-22independentSource ↗
InfiniteBenchNearing Saturation71.6%
65
2025-04-22independentSource ↗
MercuryDeprecated67.3%
10
2025-04-21independentSource ↗
SCROLLSNearing Saturation67.4%
65
2025-04-21independentSource ↗
MMLU-ProNearing Saturation77.5%
75
2025-04-21independentSource ↗
RULERNearing Saturation77.5%
84
2025-04-20independentSource ↗
TabMWPSaturated73.6%
25
2025-04-20independentSource ↗
CyberSecEvalDeprecated68.9%
10
2025-04-20independentSource ↗
SWE-bench VerifiedSaturated59.7%
25
2025-04-19independentSource ↗
MMLU-ReduxSaturated82.5%
25
2025-04-19independentSource ↗
QuALITYNearing Saturation74.1%
78
2025-04-19independentSource ↗
LongBenchNearing Saturation72.4%
65
2025-04-18independentSource ↗
SWE-benchDeprecated52%
10
2025-04-18independentSource ↗
MGSMNearing Saturation76%
65
2025-04-18independentSource ↗
AGIEvalSaturated71.2%
25
2025-04-17independentSource ↗
QASPERNearing Saturation69.9%
65
2025-04-17independentSource ↗
TruthfulQASaturated74.1%
25
2025-04-17independentSource ↗
GSM-HardDeprecated76.8%
10
2025-04-16independentSource ↗
SuperGLUESaturated84.5%
25
2025-04-16independentSource ↗
CodeXGLUEDeprecated71.9%
23
2025-04-15independentSource ↗
MT-BenchNearing Saturation8%
75
2025-04-15independentSource ↗
BIG-Bench HardSaturated81.1%
25
2025-04-15independentSource ↗
WinoGrandeSaturated86.1%
28
2025-04-14independentSource ↗
MBPP+Nearing Saturation78%
65
2025-04-14independentSource ↗
IFEvalDeprecated75.9%
25
2025-04-14independentSource ↗
GPQANearing Saturation74.1%
78
2025-04-14independentSource ↗
Needle-in-a-HaystackSaturated96.8%
44
2025-04-13independentSource ↗
HumanEval+Saturated82.6%
25
2025-04-13independentSource ↗
CommonsenseQASaturated86.9%
28
2025-04-13independentSource ↗
MBPPSaturated81.1%
25
2025-04-12independentSource ↗
GSM8KSaturated86.4%
25
2025-04-12independentSource ↗
GLUESaturated89.4%
28
2025-04-12independentSource ↗
MMLUSaturated85.9%
25
2025-04-11independentSource ↗
HumanEvalSaturated87.2%
25
2025-04-11independentSource ↗
ARC-AGISaturated77.8%
25
2025-04-11independentSource ↗
LiveBench (Continuous)Active74%
94
2025-02-01independentSource ↗
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI

client = OpenAI()
response = client.chat.completions.create(
    model="sarvam-30b",
    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 30B 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.