Sarvam AIproprietary
Released: 2025-06-15

Sarvam 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 composite
Throughput Speed
72tok/sec
Output streaming throughput
Time To First Token
340ms
Initial chunk server latency
LiveBench Contam-Free
78.4%
September 2026 suite score

Sourced Benchmark Evaluations (61)

Verified performance across authoritative benchmarks with provenance tracking.
BenchmarkStatusScoreTrustDateSource TypeProvenance
ARC-AGI-3Active56.7%
93
2025-07-20independentSource ↗
ARC-AGI-2Nearing Saturation68%
75
2025-07-13independentSource ↗
FrontierMathActive23.7%
98
2025-07-10independentSource ↗
PutnamBenchActive43.9%
90
2025-07-09independentSource ↗
SWE-LancerDeprecated55.1%
20
2025-07-07independentSource ↗
Omni-MATHDeprecated62.5%
10
2025-07-07independentSource ↗
SoohakActive64.2%
98
2025-07-05independentSource ↗
FrontierCodeActive47%
98
2025-07-05independentSource ↗
Humanity's Last ExamActive38.4%
93
2025-07-03independentSource ↗
StructFlowBenchNearing Saturation69.2%
75
2025-07-03independentSource ↗
AIMENearing Saturation80.3%
75
2025-07-03independentSource ↗
OlympiadBenchActive69.3%
80
2025-07-02independentSource ↗
MathArenaActive67.6%
99
2025-07-01independentSource ↗
FireBenchActive68.4%
95
2025-07-01independentSource ↗
Terminal-BenchActive56.7%
80
2025-07-01independentSource ↗
OpenAI-MRCRActive74.3%
80
2025-06-30independentSource ↗
SimpleQAActive46%
90
2025-06-30independentSource ↗
CodeContestsDeprecated52.7%
10
2025-06-30independentSource ↗
ComplexBenchNearing Saturation66.8%
80
2025-06-29independentSource ↗
SWE-bench ProActive50.2%
90
2025-06-29independentSource ↗
LongBench-ProActive72.6%
80
2025-06-29independentSource ↗
ZeroSCROLLSNearing Saturation70.8%
78
2025-06-28independentSource ↗
MultiPL-EDeprecated74.5%
10
2025-06-28independentSource ↗
LiveCodeBenchSaturated66.4%
39
2025-06-27independentSource ↗
AgentIFNearing Saturation55.4%
80
2025-06-27independentSource ↗
MATHSaturated82.8%
25
2025-06-27independentSource ↗
InfiniteBenchNearing Saturation75.2%
65
2025-06-27independentSource ↗
MercuryDeprecated71.3%
10
2025-06-26independentSource ↗
SCROLLSNearing Saturation70.8%
65
2025-06-26independentSource ↗
MMLU-ProNearing Saturation81.4%
75
2025-06-26independentSource ↗
RULERNearing Saturation81.4%
84
2025-06-25independentSource ↗
TabMWPSaturated77.7%
25
2025-06-25independentSource ↗
CyberSecEvalDeprecated72.9%
10
2025-06-25independentSource ↗
SWE-bench VerifiedSaturated63.2%
25
2025-06-24independentSource ↗
MMLU-ReduxSaturated86.7%
25
2025-06-24independentSource ↗
QuALITYNearing Saturation77.9%
78
2025-06-24independentSource ↗
LongBenchNearing Saturation76.1%
65
2025-06-23independentSource ↗
SWE-benchDeprecated55.1%
10
2025-06-23independentSource ↗
MGSMNearing Saturation80.3%
65
2025-06-23independentSource ↗
AGIEvalSaturated75%
25
2025-06-22independentSource ↗
QASPERNearing Saturation73.5%
65
2025-06-22independentSource ↗
TruthfulQASaturated77.9%
25
2025-06-22independentSource ↗
GSM-HardDeprecated81.1%
10
2025-06-21independentSource ↗
SuperGLUESaturated88.9%
25
2025-06-21independentSource ↗
CodeXGLUEDeprecated76.1%
23
2025-06-20independentSource ↗
MT-BenchNearing Saturation8.4%
75
2025-06-20independentSource ↗
BIG-Bench HardSaturated85.5%
25
2025-06-20independentSource ↗
WinoGrandeSaturated90.7%
28
2025-06-19independentSource ↗
MBPP+Nearing Saturation82.6%
65
2025-06-19independentSource ↗
IFEvalDeprecated79.8%
25
2025-06-19independentSource ↗
GPQANearing Saturation77.9%
78
2025-06-19independentSource ↗
Needle-in-a-HaystackSaturated97.7%
44
2025-06-18independentSource ↗
HumanEval+Saturated87.5%
25
2025-06-18independentSource ↗
CommonsenseQASaturated91.6%
28
2025-06-18independentSource ↗
MBPPSaturated85.9%
25
2025-06-17independentSource ↗
GSM8KSaturated91.3%
25
2025-06-17independentSource ↗
GLUESaturated94.2%
28
2025-06-17independentSource ↗
MMLUSaturated90.3%
25
2025-06-16independentSource ↗
HumanEvalSaturated92.3%
25
2025-06-16independentSource ↗
ARC-AGISaturated82%
25
2025-06-16independentSource ↗
LiveBench (Continuous)Active78.4%
94
2025-02-01independentSource ↗
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.