GoogleReasoning modelGemini 2.5 Flash
Best configuration ranks #312 of 610 on the BenchLeader Index at 48.8 ±3.6. Last measured 2 Sept 2026. Released 17 Apr 2025.
- Blended price
- $0.850/M
- $0.300 in · $2.50 out
- Output speed
- 193 tok/s
- First answer
- 0.43 s
- first token 0.86 s
- Context
- 1.0M
- Overall index49
- Reasoning42
- Coding43
- Agents & tools44
- Maths56
- Knowledge47
- Instruction following42
- Human preference60
- Multimodal58
- Long context55
- Composite41
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 |
|---|---|---|---|---|---|---|
| thinking | 48.1 | #334 | 211 tok/s | 17 s | $0.0009 | Agents & tools 45 · Coding 35 · Composite 46 · Instruction following 51 · Knowledge 52 · Long context 59 · Multimodal 53 · Reasoning 40 |
| defaultbest | 48.8 | #312 | 193 tok/s | 0.43 s | $0.0009 | Agents & tools 44 · Coding 43 · Composite 41 · Human preference 60 · Instruction following 42 · Knowledge 47 · Long context 55 · Maths 56 · Multimodal 58 · Reasoning 42 |
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 |
|---|---|---|---|
| Humanity's Last Exam | – | 12.1%#24 | Scale AI / CAIS |
| SimpleBench | – | 41.2%#58 | SimpleBench |
| LMArena Hard Prompts | – | 1420#130 | LMArena |
| GPQA Diamond (AA)not in index | 79.0%#197 | 68.3%#300 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 12.1%#218 | 4.7%#400 | Artificial Analysis |
| GPQA Diamond (Vals)not in index | – | 57.6%#110 | Vals AI |
| Kagi LLM Benchmark | 56.8%#60 | 44.1%#96 | Kagi LLM Benchmark |
| ARC-AGI-1 | 33.3%#140 | 33.3%#140 | ARC Prize |
| ARC-AGI-2 | 2.5%#139 | 1.7%#148 | ARC Prize |
Coding
| Benchmark | thinking | default | Source |
|---|---|---|---|
| WeirdML | 41.0%#97 | 41.0%#97 | WeirdML |
| LMArena Coding | – | 1424#157 | LMArena |
| LiveCodeBench | 46.9%#114 | 56.9%#106 | Vals AI |
| IOI | – | 2.6%#53 | Vals AI |
| Aider Polyglot | – | 55.1%#18 | Aider polyglot leaderboard |
| SWE-bench Verified (bash only) | – | 28.7%#37 | SWE-bench |
| SWE-bench Verified (any scaffold)not in index | – | 28.7%#56 | SWE-bench |
Agents & tools
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Terminal-Bench | – | 17.1%#59 | Terminal-Bench |
| APEX-Agents | – | 1.8%#61 | Mercor |
| Terminal-Bench Hard | 13.6%#194 | 12.1%#202 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 31.6%#245 | 14.9%#348 | Artificial Analysis |
| BFCL Overall | – | 56.2%#13 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | thinking | default | Source |
|---|---|---|---|
| OTIS Mock AIME | – | 73.1%#119 | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | thinking | default | Source |
|---|---|---|---|
| AA-Omniscience | -29.8#213 | -42.6#269 | Artificial Analysis |
| LegalBench | – | 83.8%#43 | Vals AI |
| CorpFin | – | 54.2%#92 | Vals AI |
| TaxEval | 70.5%#88 | 71.2%#80 | Vals AI |
| MedQA | 91.0%#42 | 86.7%#57 | Vals AI |
| PRBench Finance | – | 38.4%#27 | Scale AI SEAL |
| PRBench Legal | – | 41.0%#20 | Scale AI SEAL |
Instruction following
| Benchmark | thinking | default | Source |
|---|---|---|---|
| IFBench | 50.3%#170 | 39.0%#263 | Artificial Analysis |
Human preference
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Text | – | 1410#122 | LMArena |
Multimodal
| Benchmark | thinking | default | Source |
|---|---|---|---|
| LMArena Vision | – | 1236#58 | LMArena |
| MMMU-Pro | 69.1%#126 | 65.5%#141 | Artificial Analysis |
| VISTA | – | 49.1%#13 | Scale AI SEAL |
| MMMU (validation) | – | 79.7%#6 | MMMU |
Long context
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Fiction.LiveBench 120k | – | 68.8%#8 | Fiction.live |
| AA-LCR | 65.3%#191 | 49.9%#255 | Artificial Analysis |
Composite
| Benchmark | thinking | default | Source |
|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | 143#82 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 13.1#253 | 9.8#318 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Google Vertex (EU) | 96 tok/s | 0.50 s | $0.300 | $2.50 | 1.0M | – |
| Google Vertex (Global) (ZDR) | 73 tok/s | 0.86 s | $0.300 | $2.50 | 1.0M | – |
| Google Vertex (Global) (ZDR) | 42 tok/s | 1.26 s | $0.540 | $4.50 | 1.0M | – |
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 at $0.030 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0009 | $0.0008 | 2.0 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0051 | $0.0027 | 3.5 s |
| Code edit | 6,000 / 1,500 | $0.0056 | $0.0043 | 8.2 s |
| Agentic coding session | 60,000 / 4,000 | $0.028 | $0.016 | 21.2 s |
| Structured extraction | 2,000 / 200 | $0.0011 | $0.0007 | 1.5 s |
See also
Data as of 9 Sept 2026. Compare these configurations.