Sarvam AIproprietary
Released: 2025-03-01Bulbul v3 (Text-to-Speech)
Human-grade expressive conversational speech synthesis engine supporting regional accents, emotional inflections, and natural prosody in 15+ Indian languages.
Intelligence Index
80.5/ 100
Calibrated multi-domain compositeThroughput Speed
150tok/sec
Output streaming throughputTime To First Token
220ms
Initial chunk server latencyLiveBench Contam-Free
70.0%
September 2026 suite scoreSourced Benchmark Evaluations (51)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 49.4% | 93 | 2025-04-05 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 59.3% | 75 | 2025-03-29 | independent | Source ↗ |
| FrontierMath | Active | 16.8% | 98 | 2025-03-26 | independent | Source ↗ |
| PutnamBench | Active | 31.2% | 90 | 2025-03-25 | independent | Source ↗ |
| Omni-MATH | Deprecated | 44.4% | 10 | 2025-03-23 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 37.4% | 20 | 2025-03-23 | independent | Source ↗ |
| Soohak | Active | 45.6% | 98 | 2025-03-21 | independent | Source ↗ |
| FrontierCode | Active | 31.9% | 98 | 2025-03-21 | independent | Source ↗ |
| AIME | Nearing Saturation | 57% | 75 | 2025-03-19 | independent | Source ↗ |
| Humanity's Last Exam | Active | 33.4% | 93 | 2025-03-19 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 63% | 75 | 2025-03-19 | independent | Source ↗ |
| OlympiadBench | Active | 49.2% | 80 | 2025-03-18 | independent | Source ↗ |
| MathArena | Active | 48% | 99 | 2025-03-17 | independent | Source ↗ |
| FireBench | Active | 62.2% | 95 | 2025-03-17 | independent | Source ↗ |
| Terminal-Bench | Active | 38.5% | 80 | 2025-03-17 | independent | Source ↗ |
| CodeContests | Deprecated | 35.8% | 10 | 2025-03-16 | independent | Source ↗ |
| SimpleQA | Active | 41.9% | 90 | 2025-03-16 | independent | Source ↗ |
| SWE-bench Pro | Active | 34.1% | 90 | 2025-03-15 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 60.7% | 80 | 2025-03-15 | independent | Source ↗ |
| MultiPL-E | Deprecated | 50.6% | 10 | 2025-03-14 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 50.4% | 80 | 2025-03-13 | independent | Source ↗ |
| MATH | Saturated | 58.8% | 25 | 2025-03-13 | independent | Source ↗ |
| LiveCodeBench | Saturated | 45.1% | 39 | 2025-03-13 | independent | Source ↗ |
| Mercury | Deprecated | 48.4% | 10 | 2025-03-12 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 74.1% | 75 | 2025-03-12 | independent | Source ↗ |
| CyberSecEval | Deprecated | 49.5% | 10 | 2025-03-11 | independent | Source ↗ |
| TabMWP | Saturated | 55.2% | 25 | 2025-03-11 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 42.9% | 25 | 2025-03-10 | independent | Source ↗ |
| MMLU-Redux | Saturated | 78.9% | 25 | 2025-03-10 | independent | Source ↗ |
| SWE-bench | Deprecated | 37.4% | 10 | 2025-03-09 | independent | Source ↗ |
| MGSM | Nearing Saturation | 57% | 65 | 2025-03-09 | independent | Source ↗ |
| TruthfulQA | Saturated | 70.8% | 25 | 2025-03-08 | independent | Source ↗ |
| AGIEval | Saturated | 65.4% | 25 | 2025-03-08 | independent | Source ↗ |
| SuperGLUE | Saturated | 77.5% | 25 | 2025-03-07 | independent | Source ↗ |
| GSM-Hard | Deprecated | 57.6% | 10 | 2025-03-07 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 7.6% | 75 | 2025-03-06 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 51.7% | 23 | 2025-03-06 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 74.5% | 25 | 2025-03-06 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 56.1% | 65 | 2025-03-05 | independent | Source ↗ |
| IFEval | Deprecated | 72.6% | 25 | 2025-03-05 | independent | Source ↗ |
| WinoGrande | Saturated | 79% | 28 | 2025-03-05 | independent | Source ↗ |
| GPQA | Nearing Saturation | 70.8% | 78 | 2025-03-05 | independent | Source ↗ |
| HumanEval+ | Saturated | 59.4% | 25 | 2025-03-04 | independent | Source ↗ |
| CommonsenseQA | Saturated | 79.8% | 28 | 2025-03-04 | independent | Source ↗ |
| MBPP | Saturated | 58.3% | 25 | 2025-03-03 | independent | Source ↗ |
| GSM8K | Saturated | 64.8% | 25 | 2025-03-03 | independent | Source ↗ |
| GLUE | Saturated | 82.1% | 28 | 2025-03-03 | independent | Source ↗ |
| MMLU | Saturated | 82.1% | 25 | 2025-03-02 | independent | Source ↗ |
| ARC-AGI | Saturated | 71.4% | 25 | 2025-03-02 | independent | Source ↗ |
| HumanEval | Saturated | 62.7% | 25 | 2025-03-02 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 70% | 94 | 2025-02-01 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="bulbul-v3",
messages=[{"role": "user", "content": "Evaluate multi-step logic problem"}],
temperature=0.2
)
print(response.choices[0].message.content)Evaluation & Sourcing Notes
Scores listed for Bulbul v3 (Text-to-Speech) 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.