gemini-2.5-flash: benchmark scores, price and ranking
gemini-2.5-flash from Google DeepMind ranks #572 of 743 LLMs on the LLMs Tiger aggregate benchmark score (45.6), based on 9 benchmark results captured up to 2026-10-13.
- Organisation
- Google DeepMind
- Type
- Fast / low-cost (flash) model
- Open weights
- No
- Released
- 2025-04-17 (source: curated)
- Context window
- 1,048,576 tokens
- Cheapest API price
- $0.300 in / $2.50 out per million tokens (via deepinfra, observed 2026-10-13T19:47Z)
- Aggregate rank
- #572 of 743 (score 45.6)
Relative to other models, gemini-2.5-flash is strongest on BALROG (#10 of 34) and weakest on Terminal-Bench 2.0 (#27 of 28).
Across the 9 benchmarks where at least ten models are scored, gemini-2.5-flash is in the top 10% on 0 and in the top half on 2.
Its cheapest listed input price, $0.300 per million tokens, is lower than that of 58% of the 427 models with a listed price.
gemini-2.5-flash benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| BALROG | 33.5% | #10 of 34 | Epoch AI Benchmarking Hub | 2026-10-13 |
| DTBench | 76.5% | #63 of 146 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SimpleBench | 41.2% | #32 of 61 | Epoch AI Benchmarking Hub | 2026-10-13 |
| LMCA | 27.5% | #60 of 108 | Epoch AI Benchmarking Hub | 2026-10-13 |
| ALE-Bench | 662 | #47 of 79 | Epoch AI Benchmarking Hub | 2026-10-13 |
| CritPt | 1.1% | #69 of 115 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Vending-Bench 2 | $549 | #14 of 19 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SWE-bench Verified | 28.7% | #61 of 73 | SWE-bench Verified leaderboard | 2026-10-13 |
| Terminal-Bench 2.0 | 17.1% | #27 of 28 | Epoch AI Benchmarking Hub | 2026-10-13 |
About gemini-2.5-flash
gemini-2.5-flash is a smaller, lower-cost model from Google DeepMind. It was released on 2025-04-17, according to the developer's announcement. It accepts up to 1,048,576 tokens of input at once. The cheapest API listing we track charges $0.300 per million input tokens and $2.50 per million output tokens, through deepinfra. We hold 9 benchmark results for it across 4 areas, and none is in the top tenth of its board.
Where gemini-2.5-flash stands
On our aggregate score, gemini-2.5-flash is #572 of 743 ranked models, in the lower half. Its score is 45.6; the scale is centred on 50, so it sits 4.4 points below the typical model. That beats 23% of the ranked field. Among the 41 ranked Google DeepMind models it is #26. The next one up is Gemma 2 9B at 46.5. The next one down is Gemini 2.0 Flash at 44.6.
gemini-2.5-flash, benchmark by benchmark
BALROG puts gemini-2.5-flash at 33.5%, #10 of 34. That beats 71% of the models tested, but 34.8 points short of GPT-6 Astra (max) (68.3%). The results either side are DeepSeek-R1 (34.9%) and Claude 3.5 Sonnet (Oct 2024) (32.6%). It covers long-horizon game-playing agents. The board median is 27.6%, so gemini-2.5-flash is 5.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
On DTBench, gemini-2.5-flash reaches 76.5%. It ranks #63 of 146, ahead of 57% of the field, and 22.4 points behind Claude Opus 5.5 (max) at 98.9%. The results either side are Qwen 3.6 Flash (2026-04-16) (77.1%) and gpt-oss-120b (high) (76.3%). It covers dynamic tool-use benchmark. The board median is 68.8%, so gemini-2.5-flash is 7.7 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
gemini-2.5-flash lands below the middle on SimpleBench: 41.2%, #32 of 61, with 51% of the field scoring higher. Claude Fable 5 (max) leads at 81.9%; the closest result above is Claude 3.5 Sonnet (Oct 2024) at 41.4%. It covers trick questions that test common sense. The board median is 41.4%, so gemini-2.5-flash is 0.2 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
gemini-2.5-flash lands below the middle on LMCA: 27.5%, #60 of 108, with 55% of the field scoring higher. Claude Opus 5.5 (max) leads at 68.2%; the closest result above is Qwen3-Max-Instruct at 28.3%. It covers language-model capability assessment. The board median is 29.2%, so gemini-2.5-flash is 1.7 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
gemini-2.5-flash lands below the middle on ALE-Bench: 662, #47 of 79, with 58% of the field scoring higher. GPT-6 Astra (max) leads at 2,951; the closest result above is Qwen 3.6 Plus (2026-04-02) at 670. It covers AtCoder heuristic-optimisation contests (performance rating). The board median is 764, so gemini-2.5-flash is 102 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
gemini-2.5-flash lands below the middle on CritPt: 1.1%, #69 of 115, with 59% of the field scoring higher. GPT-5.6 Sol (max) leads at 32.3%; the closest result above is DeepSeek-R1 at 1.1%. It covers unpublished research-level physics problems. The board median is 2.9%, so gemini-2.5-flash is 1.8 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
gemini-2.5-flash lands below the middle on Vending-Bench 2: $549, #14 of 19, with 68% of the field scoring higher. Kimi K2.6 leads at $6,205; the closest result above is grok-4-1-fast-reasoning at $1,107. It covers running a vending business for a year (final money, $). The board median is $2,377, so gemini-2.5-flash is 1,828 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
SWE-bench Verified is a weak spot. gemini-2.5-flash scores 28.7%, #61 of 73, behind 82% of the models measured. The leader, Claude Opus 4.7 (max), reaches 83.5%. It covers real GitHub issues resolved. The board median is 56.4%, so gemini-2.5-flash is 27.7 points below it. Source: SWE-bench Verified leaderboard, captured 2026-10-13.
Terminal-Bench 2.0 is a weak spot. gemini-2.5-flash scores 17.1%, #27 of 28, behind 93% of the models measured. The leader, Gemini 3 Pro Preview, reaches 69.4%. It covers real terminal tasks (agent + model). The board median is 37.3%, so gemini-2.5-flash is 20.2 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Strengths and weaknesses
Software engineering: weak, ahead of 20% of the field on average across 3 benchmarks. Its best board here is ALE-Bench (#47 of 79) and its weakest is Terminal-Bench 2.0 (#27 of 28). Reasoning, maths and science: uneven, ahead of 44% of the field on average across 2 benchmarks. Its best board here is SimpleBench (#32 of 61) and its weakest is CritPt (#69 of 115). Agents and tool use: solid, ahead of 64% of the field on average across 2 benchmarks. Its best board here is BALROG (#10 of 34) and its weakest is DTBench (#63 of 146). General knowledge and chat: uneven, ahead of 35% of the field on average across 2 benchmarks. Its best board here is LMCA (#60 of 108) and its weakest is Vending-Bench 2 (#14 of 19).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so gemini-2.5-flash costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on qwen3.6-plus ($0.325) and Qwen3-Coder-480B-A35B-Instruct ($0.380).
Release history
The Google DeepMind model released before it was Gemini 2.5 Pro Preview (Mar 2025) on 2025-03-31, which scores 50.9. The next one was Gemini 2.5 Flash Preview (Apr 2025) on 2025-04-17, at 48.6.
Reading these numbers
The 9 scores on this page come from 2 sources: Epoch AI Benchmarking Hub (8), SWE-bench Verified leaderboard (1). 1 of them come from an official leaderboard and 8 from datasets or third-party runs.
Models ranked nearby
- #569 ollama/mistral-nemo:12b-instruct-2407-q4_K_M (45.6)
- #570 Llama 4 Maverick (FP8) (45.6)
- #571 mistral-large-2411 (45.6)
- #573 qwen3-14b (45.6)
- #574 Yi-1.5-6B-Chat (45.6)
- #575 qwen3-coder-next (45.5)
More from Google DeepMind
- Gemini 3.8 Flash (high) (#26, 54.0)
- Gemini 3.7 Flash (high) (#27, 53.9)
- Gemini 3 Pro Preview (#31, 53.7)
- Gemini 3 Deep Think (#60, 52.6)
- AI co-mathematician (#69, 52.3)
- Gemini 3 Flash (#120, 51.5)
Compare it on the live leaderboard → · How the aggregate score works