GoogleAuto-detectedGemini 2.5 Flash 09
Best configuration ranks #246 of 610 on the BenchLeader Index at 52.5 ±2.4 (thinking reasoning effort). Last measured 1 Sept 2026. Released 25 Sept 2025.
- Blended price
- $0.850/M
- $0.300 in · $2.50 out
- Output speed
- –
- First answer
- –
- Context
- 1M
- Overall index53
- Coding54
- Agents & tools48
- Maths48
- Knowledge53
- Instruction following53
- Multimodal58
- Long context62
- Composite49
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| thinkingbest | 52.5 | #246 | – | – | $0.0009 | Agents & tools 48 · Coding 54 · Composite 49 · Instruction following 53 · Knowledge 53 · Long context 62 · Maths 48 · Multimodal 58 |
| default | 50.7 | #279 | – | – | $0.0009 | Agents & tools 36 · Coding 52 · Composite 45 · Human preference 59 · Instruction following 45 · Knowledge 52 · Long context 56 · Maths 47 · Multimodal 57 · Reasoning 59 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Hard Prompts | – | 1420#129 | LMArena |
| GPQA Diamond (AA)not in index | 79.3%#193 | 76.6%#224 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 13.8%#205 | 8.7%#277 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | 76.5%#75 | 81.6%#62 | Vals AI |
Coding
| Benchmark | thinking | default | Source |
|---|---|---|---|
| WeirdML | – | 41.9%#92 | WeirdML |
| LMArena Coding | – | 1428#154 | LMArena |
| LiveCodeBench | 76.2%#76 | 75.1%#77 | Vals AI |
Agents & tools
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Terminal-Bench | – | 17.1%#59 | Terminal-Bench |
| Terminal-Bench Hard | 16.7%#179 | 14.4%#189 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 45.6%#203 | 28.4%#260 | Artificial Analysis |
Maths
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AIME (Vals) | 51.5%#59 | 49.8%#60 | Vals AI |
| MGSM | 89.8%#41 | 89.8%#40 | Vals AI |
Knowledge
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA-Omniscience | -36.1#238 | -39.9#254 | Artificial Analysis |
| MMLU-Pro | 83.7%#69 | 83.7%#68 | Vals AI |
| LegalBench | 82.6%#59 | 82.5%#62 | Vals AI |
| CorpFin | 59.8%#71 | 59.0%#77 | Vals AI |
| TaxEval | 72.4%#62 | 72.7%#58 | Vals AI |
| MedQA | 91.2%#40 | 91.4%#37 | Vals AI |
Instruction following
| Benchmark | thinking | default | Source |
|---|---|---|---|
| IFBench | 52.3%#159 | 43.5%#218 | Artificial Analysis |
Human preference
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Text | – | 1404#127 | LMArena |
Multimodal
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Vision | – | 1253#47 | LMArena |
| MMMU-Pro | 73.1%#98 | 70.2%#115 | Artificial Analysis |
Long context
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA-LCR | 71.0%#152 | 60.0%#219 | Artificial Analysis |
Composite
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA Intelligence Index | 15.5#214 | 12.4#267 | Artificial Analysis |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0009 | – | – |
| Summarise a 30-page report | 12,000 / 600 | $0.0051 | – | – |
| Code edit | 6,000 / 1,500 | $0.0056 | – | – |
| Agentic coding session | 60,000 / 4,000 | $0.028 | – | – |
| Structured extraction | 2,000 / 200 | $0.0011 | – | – |
See also
Data as of 9 Sept 2026. Compare these configurations.