Model comparison
DeepSeek V4 Flash vs Gemini 3 Flash Preview
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 52.3 on the Noometry Index.
Last verified . 36 shared benchmarks.
DeepSeek
GoogleSummary
- They share 36 benchmarks with published results for both. DeepSeek V4 Flash scores higher in 2 categories and Gemini 3 Flash Preview in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Flash leads 60.3 to 51.7.
- The biggest single-benchmark swing is ProofBench: 56% for DeepSeek V4 Flash and 15% for Gemini 3 Flash Preview.
- DeepSeek V4 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $0.50 / $3 for Gemini 3 Flash Preview.
- Gemini 3 Flash Preview accepts more context: 1.05M tokens versus 1M.
- DeepSeek V4 Flash has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4 Flash | Gemini 3 Flash Preview | |
|---|---|---|
| Provider | DeepSeek | |
| Noometry Index | 53.6 | 52.3 |
| Released | 2026-04-24 | 2025-12-17 |
| Weights | Open | Proprietary |
| Context window | 1M | 1.05M |
| Max output | 393K | 66K |
| Input $ / M tokens | $0.15 | $0.50 |
| Output $ / M tokens | $0.60 | $3 |
| Results tracked | 41 | 59 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Gemini 3 Flash Preview leads
DeepSeek V4 Flash: 47.9 (#59), Gemini 3 Flash Preview: 50.9 (#42)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| LMArena WebDev | 1582 | 1439 |
| WeirdML | 63% | 61.6% |
| LMArena Coding | 1457 | 1460 |
| ALE-Bench | 1,306 | 1,367 |
| SWE-bench Verified | — | 75.4% |
| FrontierCode | 18.8% | — |
| SWE-bench Verified (bash only) | — | 75.8% |
| SWE-bench Multilingual | — | 72.7% |
| SciCode | 49.9% | — |
| GSO | — | 9.8% |
Agentic & Tool Use Not comparable
DeepSeek V4 Flash: —, Gemini 3 Flash Preview: 38.7 (#29)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| Terminal-Bench | — | 64.3% |
| τ²-bench Airline | — | 82.5% |
| τ²-bench Banking | — | 27.3% |
| τ²-bench Retail | — | 76.8% |
| τ²-bench Telecom | — | 91.2% |
| DeepResearch Bench | — | 49.8% |
| BALROG | — | 48.1% |
| GDP.pdf | — | 10% |
| LMArena Search | — | 1198 |
| Vending-Bench 2 | — | 3,635 |
Reasoning DeepSeek V4 Flash leads
DeepSeek V4 Flash: 53.7 (#30), Gemini 3 Flash Preview: 49.2 (#37)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| ARC-AGI-2 | 61.4% | 33.6% |
| SimpleBench | 61.1% | 61.1% |
| NYT Connections (extended) | 89.6% | 83.1% |
| ARC-AGI-1 | 89% | 84.7% |
| Chess Puzzles | 33% | 40% |
| LMArena Hard Prompts | 1444 | 1465 |
| Mystery Game Puzzles | 34% | 26% |
| DTBench | 90.9% | 89.1% |
| LMCA | 41.7% | 43.1% |
| Epoch Capabilities Index | 154.49 | 151.8 |
| Kagi LLM Benchmark | 52.2% | — |
| CritPt | 16.6% | — |
| ForecastBench | — | 58.5 |
Math DeepSeek V4 Flash leads
DeepSeek V4 Flash: 60.3 (#37), Gemini 3 Flash Preview: 51.7 (#55)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| FrontierMath (Tiers 1-3) | 57.5% | 51.2% |
| FrontierMath Tier 4 | 24.4% | 17.1% |
| MathArena Final-Answer Competitions | 76.5% | 67.6% |
| OTIS Mock AIME 2024-2025 | 94.4% | 95.6% |
| ProofBench | 56% | 15% |
| LMArena Math | 1427 | 1473 |
| FrontierMath (Feb 2025 set) | — | 35.6% |
| FrontierMath Tier 4 (v1) | — | 4.2% |
Knowledge Gemini 3 Flash Preview leads
DeepSeek V4 Flash: 55.4 (#48), Gemini 3 Flash Preview: 58.8 (#33)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| GPQA Diamond | 91% | 89.4% |
| SimpleQA Verified | 33.6% | 66.8% |
| LMArena Expert | 1441 | 1462 |
| Vectara Hallucination Rate | — | 13.5% |
Multimodal Not comparable
DeepSeek V4 Flash: —, Gemini 3 Flash Preview: 45.5 (#16)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| LMArena Vision | — | 1285 |
| GeoBench | — | 88% |
| VPCT | — | 72.6% |
| Blueprint-Bench 2 | — | 0% |
| LMArena Document | — | 1413 |
Multilingual Gemini 3 Flash Preview leads
DeepSeek V4 Flash: 53.0 (#72), Gemini 3 Flash Preview: 55.7 (#27)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| LMArena Non-English | 1420 | 1458 |
| LMArena Chinese | 1468 | 1511 |
| LMArena French | 1439 | 1477 |
| LMArena German | 1418 | 1497 |
| LMArena Japanese | 1406 | 1489 |
| LMArena Korean | 1384 | 1443 |
| LMArena Russian | 1428 | 1480 |
| LMArena Spanish | 1436 | 1469 |
Instruction Following Too close to call
DeepSeek V4 Flash: 74.9 (#81), Gemini 3 Flash Preview: 75.7 (#56)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| LMArena Instruction Following | 1421 | 1437 |
Long Context Too close to call
DeepSeek V4 Flash: 43.8 (#85), Gemini 3 Flash Preview: 44.4 (#67)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| LMArena Longer Query | 1434 | 1452 |
Writing & Preference Gemini 3 Flash Preview leads
DeepSeek V4 Flash: 63.8 (#61), Gemini 3 Flash Preview: 65.5 (#45)
| Benchmark | DeepSeek V4 Flash | Gemini 3 Flash Preview |
|---|---|---|
| LMArena Text | 1432 | 1466 |
| LMArena Creative Writing | 1403 | 1457 |
| LMArena Multi-Turn | 1449 | 1471 |
| EQ-Bench Creative Writing | 1559 | — |
Frequently asked questions
Is DeepSeek V4 Flash better than Gemini 3 Flash Preview?
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 52.3 on the Noometry Index.
Which is cheaper, DeepSeek V4 Flash or Gemini 3 Flash Preview?
DeepSeek V4 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; Gemini 3 Flash Preview lists at $0.50 and $3.
Is DeepSeek V4 Flash or Gemini 3 Flash Preview better for coding?
Gemini 3 Flash Preview scores higher on coding benchmarks: 50.9 versus 47.9 in the Noometry coding category.
Which has the bigger context window?
Gemini 3 Flash Preview does, with 1.05M tokens against 1M.
How many benchmarks do DeepSeek V4 Flash and Gemini 3 Flash Preview share?
36 benchmarks have published results for both models. DeepSeek V4 Flash has 41 scored results on Noometry and Gemini 3 Flash Preview has 59.