Gemini 2.5 Flash
Google · released Jun 17, 2025
Gemini 2.5 Flash is Google's state-of-the-art workhorse model, specifically designed for advanced reasoning, coding, mathematics, and scientific tasks. It includes built-in "thinking" capabilities, enabling it to provide responses with greater...
Specification
- Context window
- 1.05M
- Max output
- 66K
- Knowledge cutoff
- Jan 31, 2025
- Parameters
- Undisclosed
- Licence
- Proprietary
- Serving providers
- 2
- Moderated
- No
- Uptime
- 100.0%
Intelligence
9.9
42th percentile
Coding
—
Coding Index
Agentic
—
Agentic Index
Output speed
155 t/s
Median across providers
Latency
459ms
Time to first token
Cost per task
$0.22
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- MMLU-Pro80.9%
- GPQA Diamond68.3%
- AIME 202560.3%
- AA-LCR (long context)49.9%
- LiveCodeBench49.5%
- IFBench39.0%
- τ²-bench (Telecom)14.9%
- Terminal-Bench Hard12.1%
- Humanity's Last Exam4.7%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| MMLU-Pro | 80.9% |
| GPQA Diamond | 68.3% |
| AIME 2025 | 60.3% |
| AA-LCR (long context) | 49.9% |
| LiveCodeBench | 49.5% |
| IFBench | 39.0% |
| τ²-bench (Telecom) | 14.9% |
| Terminal-Bench Hard | 12.1% |
| Humanity's Last Exam | 4.7% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- 10.3
- GPT-4.1 MiniOpenAI10.2
- Gemini 2.5 FlashGoogle9.9
- North Mini Code (free)Cohere9.9
- 9.9
- R1 Distill Llama 70BDeepSeek9.8
- 9.6
- 9.6
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| Nemotron 3 Nano Omni (free) | 10.3 |
| GPT-4.1 Mini | 10.2 |
| Gemini 2.5 Flash | 9.9 |
| North Mini Code (free) | 9.9 |
| Qwen3 VL 235B A22B Instruct | 9.9 |
| R1 Distill Llama 70B | 9.8 |
| Qwen3 Next 80B A3B Instruct | 9.6 |
| Qwen3 Coder 30B A3B Instruct | 9.6 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.3
- Output / 1M tokens
- $2.5
- Cached input / 1M
- $0.03
- Blended 3:1
- $0.85
One task, estimated
$0.22
- Input tokens
- 50,000
- Output tokens
- 80,000
- Profile
- Reasoning
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 1,076
- Win rate
- 45.6%
- Strongest at
- Websites
- Tournaments
- 5,495
Elo by category
Dot position on a 1,000–1,100 scale · Elo has no meaningful zero
- 1,076
- 1,061
- 1,021
- 1,008
- 1,007
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| Websites | 1076 | 45.6% |
| Data visualisation | 1061 | 46.8% |
| 3D scenes | 1021 | 44.4% |
| Game development | 1008 | 43.6% |
| UI components | 1007 | 43.6% |