Google DeepMindproprietary
Released: 2026-07-21Gemini 3.5 Flash-Lite
Google ultra-high-speed foundation model designed for large-scale automation at 220 tokens/sec.
Intelligence Index
87.5/ 100
Calibrated multi-domain compositeThroughput Speed
220tok/sec
Output streaming throughputTime To First Token
190ms
Initial chunk server latencyLiveBench Contam-Free
73.8%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning83.5%
Coding & Repo Repair74.0%
Mathematics (AIME)85.2%
Data Analysis & Tables71.0%
Language & Comprehension67.5%
Instruction Following61.5%
Sourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 55.6% | 93 | 2026-08-25 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 66.7% | 75 | 2026-08-18 | independent | Source ↗ |
| FrontierMath | Active | 23.9% | 98 | 2026-08-15 | independent | Source ↗ |
| PutnamBench | Active | 44.3% | 90 | 2026-08-14 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 50.3% | 20 | 2026-08-12 | independent | Source ↗ |
| Omni-MATH | Deprecated | 63% | 10 | 2026-08-12 | independent | Source ↗ |
| Soohak | Active | 64.8% | 98 | 2026-08-10 | independent | Source ↗ |
| FrontierCode | Active | 42.9% | 98 | 2026-08-10 | independent | Source ↗ |
| Humanity's Last Exam | Active | 37.6% | 93 | 2026-08-08 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 52.3% | 75 | 2026-08-08 | independent | Source ↗ |
| AIME | Nearing Saturation | 80.9% | 75 | 2026-08-08 | independent | Source ↗ |
| OlympiadBench | Active | 69.9% | 80 | 2026-08-07 | independent | Source ↗ |
| MathArena | Active | 68.2% | 99 | 2026-08-06 | independent | Source ↗ |
| FireBench | Active | 51.7% | 95 | 2026-08-06 | independent | Source ↗ |
| Terminal-Bench | Active | 51.8% | 80 | 2026-08-06 | independent | Source ↗ |
| OpenAI-MRCR | Active | 73.5% | 80 | 2026-08-05 | independent | Source ↗ |
| SimpleQA | Active | 45.5% | 90 | 2026-08-05 | independent | Source ↗ |
| CodeContests | Deprecated | 48.1% | 10 | 2026-08-05 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 50.4% | 80 | 2026-08-04 | independent | Source ↗ |
| SWE-bench Pro | Active | 45.9% | 90 | 2026-08-04 | independent | Source ↗ |
| LongBench-Pro | Active | 71.8% | 80 | 2026-08-04 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 70% | 78 | 2026-08-03 | independent | Source ↗ |
| MultiPL-E | Deprecated | 68.1% | 10 | 2026-08-03 | independent | Source ↗ |
| LiveCodeBench | Saturated | 60.7% | 39 | 2026-08-02 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 41.8% | 80 | 2026-08-02 | independent | Source ↗ |
| MATH | Saturated | 83.5% | 25 | 2026-08-02 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 74.4% | 65 | 2026-08-02 | independent | Source ↗ |
| Mercury | Deprecated | 65.1% | 10 | 2026-08-01 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 70% | 65 | 2026-08-01 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 80.5% | 75 | 2026-08-01 | independent | Source ↗ |
| RULER | Nearing Saturation | 80.5% | 84 | 2026-07-31 | independent | Source ↗ |
| TabMWP | Saturated | 78.4% | 25 | 2026-07-31 | independent | Source ↗ |
| CyberSecEval | Deprecated | 66.6% | 10 | 2026-07-31 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 57.7% | 25 | 2026-07-30 | independent | Source ↗ |
| MMLU-Redux | Saturated | 85.8% | 25 | 2026-07-30 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 77% | 78 | 2026-07-30 | independent | Source ↗ |
| LongBench | Nearing Saturation | 75.3% | 65 | 2026-07-29 | independent | Source ↗ |
| SWE-bench | Deprecated | 50.3% | 10 | 2026-07-29 | independent | Source ↗ |
| MGSM | Nearing Saturation | 80.9% | 65 | 2026-07-29 | independent | Source ↗ |
| AGIEval | Saturated | 73.5% | 25 | 2026-07-28 | independent | Source ↗ |
| QASPER | Nearing Saturation | 72.6% | 65 | 2026-07-28 | independent | Source ↗ |
| TruthfulQA | Saturated | 77% | 25 | 2026-07-28 | independent | Source ↗ |
| GSM-Hard | Deprecated | 81.8% | 10 | 2026-07-27 | independent | Source ↗ |
| SuperGLUE | Saturated | 87.2% | 25 | 2026-07-27 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 69.6% | 23 | 2026-07-26 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 6.3% | 75 | 2026-07-26 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 83.8% | 25 | 2026-07-26 | independent | Source ↗ |
| WinoGrande | Saturated | 88.9% | 28 | 2026-07-25 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 75.5% | 65 | 2026-07-25 | independent | Source ↗ |
| IFEval | Deprecated | 60.3% | 25 | 2026-07-25 | independent | Source ↗ |
| GPQA | Nearing Saturation | 77% | 78 | 2026-07-25 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 97.5% | 44 | 2026-07-24 | independent | Source ↗ |
| HumanEval+ | Saturated | 79.9% | 25 | 2026-07-24 | independent | Source ↗ |
| CommonsenseQA | Saturated | 89.8% | 28 | 2026-07-24 | independent | Source ↗ |
| MBPP | Saturated | 78.4% | 25 | 2026-07-23 | independent | Source ↗ |
| GSM8K | Saturated | 92% | 25 | 2026-07-23 | independent | Source ↗ |
| GLUE | Saturated | 92.3% | 28 | 2026-07-23 | independent | Source ↗ |
| MMLU | Saturated | 89.3% | 25 | 2026-07-22 | independent | Source ↗ |
| HumanEval | Saturated | 84.4% | 25 | 2026-07-22 | independent | Source ↗ |
| ARC-AGI | Saturated | 80.4% | 25 | 2026-07-22 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 73.8% | 94 | 2025-02-01 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gemini-3-5-flash-lite",
messages=[{"role": "user", "content": "Evaluate multi-step logic problem"}],
temperature=0.2
)
print(response.choices[0].message.content)Evaluation & Sourcing Notes
Scores listed for Gemini 3.5 Flash-Lite 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.