Gemini 2.5 Pro vs Gemini 3 Flash Preview
Updated October 21, 2026. 37 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Gemini 3 Flash Preview for coding: it scores 75.8 on the Vibe Coding Index against 37.0 for Gemini 2.5 Pro, though on 1 benchmark against 2.
Pick Gemini 2.5 Pro when what matters is CaseLaw (Vals AI) or MortgageTax (Vals AI), where it leads.
- Shared results
- 37
- Gemini 2.5 Pro only
- 50
- Gemini 3 Flash Preview only
- 32
- Like-for-like categories
- 8 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, four public coding benchmarks
Gemini 3 Flash Preview
Gemini 3 Flash Preview scores 75.8 against 37.0.
2 and 1 benchmarks behind the scores
Intelligence
AA Intelligence Index
Gemini 3 Flash Preview
Gemini 3 Flash Preview leads, 26.3 against 16.1.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Gemini 2.5 Pro has a AA Agentic Index result, 1.6.
Directional only
Reasoning
GPQA Diamond
No call
Only Gemini 3 Flash Preview has a GPQA Diamond result, 87.9%.
Directional only
Math
AIME
No call
Only Gemini 3 Flash Preview has a AIME result, 95.6%.
Directional only
Knowledge
MMLU Pro
No call
Only Gemini 3 Flash Preview has a MMLU Pro result, 88.6%.
Directional only
Web design
Design Arena: websites
Gemini 3 Flash Preview
Gemini 3 Flash Preview leads, 1196 against 1167.
Same board, both measured
Long documents
Listed context window
No call
Both take 1.05M tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemini 3 Flash Preview
Gemini 2.5 Pro $0.0063, Gemini 3 Flash Preview $0.0020 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
Gemini 3 Flash Preview
Gemini 2.5 Pro $0.64, Gemini 3 Flash Preview $0.37.
Published attempt cost and pass rate
At a glance
- Coding index
- Gemini 2.5 Pro 37.0 from 2 benchmarks, Gemini 3 Flash Preview 75.8 from 1 benchmark.
- Widest category
- Reasoning, 19.2 points apart on average over 5 shared benchmarks.
- Widest single gap
- ARC-AGI: Gemini 2.5 Pro 41.0% against Gemini 3 Flash Preview 84.7%, 43.7 points apart on a 0 to 100 scale.
- Output price
- Gemini 2.5 Pro $10.00 against Gemini 3 Flash Preview $3.00 per million output tokens, 3.3x apart.
- Context window
- Both take 1.05M tokens.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Gemini 2.5 Pro | Bars | Gemini 3 Flash Preview | Ties |
|---|---|---|---|---|
| Coding | 0 | 8 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 3 | ||
| Reasoning | 0 | 5 | ||
| Math | 0 | 5 | ||
| Knowledge | 2 | 3 | ||
| Web design | 0 | 4 | ||
| General | 1 | 5 | ||
| All shared boards | 3 | 34 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index08
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| SWE-bench Verified | 53.6% | 75.8% | Gemini 3 Flash Preview |
| LMArena WebDev | 1226 | Not measured | Only Gemini 2.5 Pro measured |
| AA Coding Index | 33.3 | Not measured | Only Gemini 2.5 Pro measured |
| LiveCodeBench (Vals AI) | Not measured | 85.6% | Only Gemini 3 Flash Preview measured |
| SWE-bench Verified (Vals AI) | 54.4% | 75.0% | Gemini 3 Flash Preview |
| Vibe Code Bench (Vals AI) | 0.4% | 20.2% | Gemini 3 Flash Preview |
| Code migration (Vals AI) | Not measured | 6.4% | Only Gemini 3 Flash Preview measured |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only Gemini 3 Flash Preview measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 53.9% | Only Gemini 3 Flash Preview measured |
| LiveCodeBench (Artificial Analysis run) | 80.1% | 90.8% | Gemini 3 Flash Preview |
| SciCode | 46.3% | Not measured | Only Gemini 2.5 Pro measured |
| Terminal-Bench Hard (Artificial Analysis run) | 26.5% | 38.6% | Gemini 3 Flash Preview |
| Terminal-Bench | 32.6% | 64.3% | Gemini 3 Flash Preview |
| GSO-Bench | 3.9% | 9.8% | Gemini 3 Flash Preview |
| WeirdML | 54.0% | 61.6% | Gemini 3 Flash Preview |
| ALE-Bench | Not measured | 1367.2 | Only Gemini 3 Flash Preview measured |
| LMArena Text: coding | 1452 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: software and IT | 1459 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena WebDev: HTML | 1235 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena WebDev: from an image | 1285 | Not measured | Only Gemini 2.5 Pro measured |
| Neither model has a result on the other 18 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 16.1 | 26.3 | Gemini 3 Flash Preview |
| Epoch Capabilities Index | 145.1 | Not measured | Only Gemini 2.5 Pro measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| AA Agentic Index | 1.6 | Not measured | Only Gemini 2.5 Pro measured |
| Design Arena: agents building web apps | Not measured | 1111 | Only Gemini 3 Flash Preview measured |
| Design Arena: agents building full-stack apps | Not measured | 1063 | Only Gemini 3 Flash Preview measured |
| Design Arena: agents building mobile apps | Not measured | 1090 | Only Gemini 3 Flash Preview measured |
| Design Arena: agents building Android apps | Not measured | 1056 | Only Gemini 3 Flash Preview measured |
| Design Arena: agents building Godot games | Not measured | 1161 | Only Gemini 3 Flash Preview measured |
| Design Arena: PowerPoint slides | Not measured | 1036 | Only Gemini 3 Flash Preview measured |
| Finance Agent (Vals AI) | Not measured | 42.5% | Only Gemini 3 Flash Preview measured |
| Tax Agent Bench (Vals AI) | Not measured | 11.8% | Only Gemini 3 Flash Preview measured |
| Legal Research Bench (Vals AI) | Not measured | 18.3% | Only Gemini 3 Flash Preview measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 26.0% | Only Gemini 3 Flash Preview measured |
| τ²-Bench Telecom | 54.1% | 80.4% | Gemini 3 Flash Preview |
| GDPval | 23.3% | Not measured | Only Gemini 2.5 Pro measured |
| BALROG | Not measured | 48.1% | Only Gemini 3 Flash Preview measured |
| DeepResearch Bench | 41.5% | 49.8% | Gemini 3 Flash Preview |
| Vending-Bench 2 | $574 | $3,635 | Gemini 3 Flash Preview |
| Neither model has a result on the other 23 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 87.9% | Only Gemini 3 Flash Preview measured |
| MMMU Pro (Vals AI) | Not measured | 87.6% | Only Gemini 3 Flash Preview measured |
| GPQA Diamond (Artificial Analysis run) | 84.4% | 89.8% | Gemini 3 Flash Preview |
| Humanity's Last Exam (Artificial Analysis run) | 22.5% | 36.6% | Gemini 3 Flash Preview |
| GPQA Diamond (Epoch AI run) | 85.3% | 89.4% | Gemini 3 Flash Preview |
| ARC-AGI | 41.0% | 84.7% | Gemini 3 Flash Preview |
| ARC-AGI-2 | 4.9% | 33.6% | Gemini 3 Flash Preview |
| SimpleBench | Not measured | 61.1% | Only Gemini 3 Flash Preview measured |
| Chess puzzles | Not measured | 39.0% | Only Gemini 3 Flash Preview measured |
| Mystery game puzzles | Not measured | 26.0% | Only Gemini 3 Flash Preview measured |
| VPCT | Not measured | 72.6% | Only Gemini 3 Flash Preview measured |
| LMArena Text: hard prompts | 1455 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: expert | 1452 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: hard prompts in English | 1453 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: diagrams | 1274 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: homework | 1281 | Not measured | Only Gemini 2.5 Pro measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| AIME | Not measured | 95.6% | Only Gemini 3 Flash Preview measured |
| MGSM (Vals AI) | Not measured | 93.3% | Only Gemini 3 Flash Preview measured |
| AA Math Index | 87.7 | 97.0 | Gemini 3 Flash Preview |
| AIME 2025 (Artificial Analysis run) | 87.7% | 97.0% | Gemini 3 Flash Preview |
| OTIS Mock AIME | 84.7% | 95.6% | Gemini 3 Flash Preview |
| FrontierMath, tiers 1 to 3 | 24.6% | 51.2% | Gemini 3 Flash Preview |
| FrontierMath, tier 4 | 0.0% | 17.1% | Gemini 3 Flash Preview |
| ProofBench | Not measured | 15.0% | Only Gemini 3 Flash Preview measured |
| LMArena Text: math | 1450 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: mathematical work | 1454 | Not measured | Only Gemini 2.5 Pro measured |
| Neither model has a result on the other 3 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index23
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 88.6% | Only Gemini 3 Flash Preview measured |
| LegalBench (Vals AI) | Not measured | 86.9% | Only Gemini 3 Flash Preview measured |
| CaseLaw (Vals AI) | 63.9% | 55.8% | Gemini 2.5 Pro |
| CorpFin (Vals AI) | 60.8% | 66.4% | Gemini 3 Flash Preview |
| TaxEval (Vals AI) | Not measured | 73.9% | Only Gemini 3 Flash Preview measured |
| MortgageTax (Vals AI) | 68.9% | 68.7% | Gemini 2.5 Pro |
| MedQA (Vals AI) | Not measured | 95.8% | Only Gemini 3 Flash Preview measured |
| MedCode (Vals AI) | 50.6% | 55.9% | Gemini 3 Flash Preview |
| MMLU Pro (Artificial Analysis run) | 86.2% | 89.0% | Gemini 3 Flash Preview |
| SimpleQA Verified | Not measured | 66.8% | Only Gemini 3 Flash Preview measured |
| GeoBench | Not measured | 88.0% | Only Gemini 3 Flash Preview measured |
| LMArena Text: science | 1480 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: medicine and healthcare | 1476 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: legal and government | 1480 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: business and finance | 1439 | Not measured | Only Gemini 2.5 Pro measured |
| Neither model has a result on the other 2 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| Design Arena: websites | 1167 | 1196 | Gemini 3 Flash Preview |
| Design Arena: UI components | 1139 | Not measured | Only Gemini 2.5 Pro measured |
| Design Arena: game dev | 1112 | 1166 | Gemini 3 Flash Preview |
| Design Arena: data viz | 1222 | Not measured | Only Gemini 2.5 Pro measured |
| Design Arena: 3D | 1080 | 1184 | Gemini 3 Flash Preview |
| Design Arena: all code categories | 1152 | 1189 | Gemini 3 Flash Preview |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index15
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 41.9% | 51.8% | Gemini 3 Flash Preview |
| MedScribe (Vals AI) | 73.5% | 69.9% | Gemini 2.5 Pro |
| Vals Multimodal Index | Not measured | 52.2% | Only Gemini 3 Flash Preview measured |
| IFBench | 48.7% | 78.0% | Gemini 3 Flash Preview |
| Long context reasoning (AA-LCR) | 69.0% | 78.0% | Gemini 3 Flash Preview |
| DTBench | 82.4% | 89.1% | Gemini 3 Flash Preview |
| LMCA | 34.8% | 43.1% | Gemini 3 Flash Preview |
| Creative writing (Lech Mazur) | 83.8% | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text | 1458 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: instruction following | 1437 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: creative writing | 1454 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision | 1263 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: English | 1458 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: non-English | 1452 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: Chinese | 1507 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: German | 1487 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: Japanese | 1461 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: Korean | 1434 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: French | 1473 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: Spanish | 1474 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: Russian | 1461 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: Polish | 1479 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: multi-turn | 1453 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: longer queries | 1449 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: writing and language | 1454 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Text: entertainment and media | 1441 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: English | 1261 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: Chinese | 1300 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: reading text | 1270 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: humor | 1262 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: captioning | 1304 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: recognizing things | 1283 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Vision: creative writing | 1265 | Not measured | Only Gemini 2.5 Pro measured |
| LMArena Document | 1421 | Not measured | Only Gemini 2.5 Pro measured |
| Neither model has a result on the other 12 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Gemini 2.5 Pro | Gemini 3 Flash Preview | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Gemini 3 Flash Preview costs less
Repository review
50K input + 3K output tokens
Gemini 3 Flash Preview costs less
Agent loop
200K input + 10K output tokens
Gemini 3 Flash Preview costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Google Gemini 2.5 Pro | Google Gemini 3 Flash Preview |
|---|---|---|
| Maker | ||
| Weights | No public weights listed | No public weights listed |
| Released | June 17, 2025 | December 17, 2025 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $1.25 | $0.50 |
| Output $/Mtok | $10.00 | $3.00 |
| Coding index | 37.0 | 75.8 |
| Independent measurements behind it | 2 | 1 |
| Cost per solved task | $0.64 | $0.37 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Gemini 2.5 Pro better than Gemini 3 Flash Preview for coding?
Pick Gemini 3 Flash Preview for coding: it scores 75.8 on the Vibe Coding Index against 37.0 for Gemini 2.5 Pro, though on 1 benchmark against 2. Pick Gemini 2.5 Pro when what matters is CaseLaw (Vals AI) or MortgageTax (Vals AI), where it leads.
Which is cheaper, Gemini 2.5 Pro or Gemini 3 Flash Preview?
Gemini 2.5 Pro costs $1.25 per million input tokens and $10.00 per million output tokens; Gemini 3 Flash Preview costs $0.50 and $3.00. Per solved SWE-bench Verified task, Gemini 2.5 Pro costs $0.64 and Gemini 3 Flash Preview $0.37.
Which has the bigger context window, Gemini 2.5 Pro or Gemini 3 Flash Preview?
Neither: both take 1.05M tokens.
How many benchmarks is the Gemini 2.5 Pro vs Gemini 3 Flash Preview comparison based on?
37 benchmarks have a result for both models as of October 21, 2026. Gemini 2.5 Pro's coding index rests on 2 independent benchmarks and Gemini 3 Flash Preview's on 1. Where only one model has a result, the page says so instead of naming a winner.