Google DeepMindproprietary
Released: 2025-02-10Gemini 2.5 Flash (Thinking)
Foundation model evaluated on LiveBench September 2026 by Google DeepMind.
Intelligence Index
88.0/ 100
Calibrated multi-domain compositeThroughput Speed
145tok/sec
Output streaming throughputTime To First Token
210ms
Initial chunk server latencyLiveBench Contam-Free
79.8%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning85.2%
Coding & Repo Repair80.5%
Mathematics (AIME)84.8%
Data Analysis & Tables76.5%
Language & Comprehension75.0%
Instruction Following77.0%
Sourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 55.4% | 93 | 2025-03-17 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 66.5% | 75 | 2025-03-10 | independent | Source ↗ |
| FrontierMath | Active | 24.4% | 98 | 2025-03-07 | independent | Source ↗ |
| PutnamBench | Active | 45.2% | 90 | 2025-03-06 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 56.1% | 20 | 2025-03-04 | independent | Source ↗ |
| Omni-MATH | Deprecated | 64.4% | 10 | 2025-03-04 | independent | Source ↗ |
| Soohak | Active | 66.1% | 98 | 2025-03-02 | independent | Source ↗ |
| FrontierCode | Active | 47.8% | 98 | 2025-03-02 | independent | Source ↗ |
| Humanity's Last Exam | Active | 37.5% | 93 | 2025-02-28 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 65.5% | 75 | 2025-02-28 | independent | Source ↗ |
| AIME | Nearing Saturation | 82.6% | 75 | 2025-02-28 | independent | Source ↗ |
| OlympiadBench | Active | 71.3% | 80 | 2025-02-27 | independent | Source ↗ |
| MathArena | Active | 69.6% | 99 | 2025-02-26 | independent | Source ↗ |
| FireBench | Active | 64.7% | 95 | 2025-02-26 | independent | Source ↗ |
| Terminal-Bench | Active | 57.7% | 80 | 2025-02-26 | independent | Source ↗ |
| OpenAI-MRCR | Active | 73.9% | 80 | 2025-02-25 | independent | Source ↗ |
| SimpleQA | Active | 45.8% | 90 | 2025-02-25 | independent | Source ↗ |
| CodeContests | Deprecated | 53.6% | 10 | 2025-02-25 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 63.1% | 80 | 2025-02-24 | independent | Source ↗ |
| SWE-bench Pro | Active | 51.2% | 90 | 2025-02-24 | independent | Source ↗ |
| LongBench-Pro | Active | 72.2% | 80 | 2025-02-24 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 70.4% | 78 | 2025-02-23 | independent | Source ↗ |
| MultiPL-E | Deprecated | 75.9% | 10 | 2025-02-23 | independent | Source ↗ |
| LiveCodeBench | Saturated | 67.6% | 39 | 2025-02-22 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 52.4% | 80 | 2025-02-22 | independent | Source ↗ |
| MATH | Saturated | 85.3% | 25 | 2025-02-22 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 74.8% | 65 | 2025-02-22 | independent | Source ↗ |
| Mercury | Deprecated | 72.6% | 10 | 2025-02-21 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 70.4% | 65 | 2025-02-21 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 81% | 75 | 2025-02-21 | independent | Source ↗ |
| RULER | Nearing Saturation | 81% | 84 | 2025-02-20 | independent | Source ↗ |
| TabMWP | Saturated | 80% | 25 | 2025-02-20 | independent | Source ↗ |
| CyberSecEval | Deprecated | 74.3% | 10 | 2025-02-20 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 64.4% | 25 | 2025-02-19 | independent | Source ↗ |
| MMLU-Redux | Saturated | 86.2% | 25 | 2025-02-19 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 77.4% | 78 | 2025-02-19 | independent | Source ↗ |
| LongBench | Nearing Saturation | 75.7% | 65 | 2025-02-18 | independent | Source ↗ |
| SWE-bench | Deprecated | 56.1% | 10 | 2025-02-18 | independent | Source ↗ |
| MGSM | Nearing Saturation | 82.6% | 65 | 2025-02-18 | independent | Source ↗ |
| AGIEval | Saturated | 73.3% | 25 | 2025-02-17 | independent | Source ↗ |
| QASPER | Nearing Saturation | 73% | 65 | 2025-02-17 | independent | Source ↗ |
| TruthfulQA | Saturated | 77.4% | 25 | 2025-02-17 | independent | Source ↗ |
| GSM-Hard | Deprecated | 83.5% | 10 | 2025-02-16 | independent | Source ↗ |
| SuperGLUE | Saturated | 86.9% | 25 | 2025-02-16 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 77.6% | 23 | 2025-02-15 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 7.9% | 75 | 2025-02-15 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 83.5% | 25 | 2025-02-15 | independent | Source ↗ |
| WinoGrande | Saturated | 88.6% | 28 | 2025-02-14 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 84.2% | 65 | 2025-02-14 | independent | Source ↗ |
| IFEval | Deprecated | 75.5% | 25 | 2025-02-14 | independent | Source ↗ |
| GPQA | Nearing Saturation | 77.4% | 78 | 2025-02-14 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 97.6% | 44 | 2025-02-13 | independent | Source ↗ |
| HumanEval+ | Saturated | 89.1% | 25 | 2025-02-13 | independent | Source ↗ |
| CommonsenseQA | Saturated | 89.5% | 28 | 2025-02-13 | independent | Source ↗ |
| MBPP | Saturated | 87.5% | 25 | 2025-02-12 | independent | Source ↗ |
| GSM8K | Saturated | 94% | 25 | 2025-02-12 | independent | Source ↗ |
| GLUE | Saturated | 92% | 28 | 2025-02-12 | independent | Source ↗ |
| MMLU | Saturated | 89.8% | 25 | 2025-02-11 | independent | Source ↗ |
| HumanEval | Saturated | 94.1% | 25 | 2025-02-11 | independent | Source ↗ |
| ARC-AGI | Saturated | 80.1% | 25 | 2025-02-11 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 79.8% | 94 | 2025-02-01 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="gemini-2-5-flash",
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 2.5 Flash (Thinking) 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.