Model comparison
DeepSeek V4 Pro vs Gemini 2.5 Flash
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 39.3 on the Noometry Index.
Last verified . 31 shared benchmarks.
DeepSeek
GoogleSummary
- They share 31 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 8 categories and Gemini 2.5 Flash in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Pro leads 56.5 to 18.1.
- The biggest single-benchmark swing is ARC-AGI-2: 61.3% for DeepSeek V4 Pro and 2.5% for Gemini 2.5 Flash.
- Gemini 2.5 Flash is cheaper at $0.30 / $2.50 per million input/output tokens, against $0.66 / $1.98 for DeepSeek V4 Pro.
- Gemini 2.5 Flash accepts more context: 1.05M tokens versus 1M.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4 Pro | Gemini 2.5 Flash | |
|---|---|---|
| Provider | DeepSeek | |
| Noometry Index | 54.3 | 39.3 |
| Released | 2026-04-24 | 2025-04-17 |
| Weights | Open | Proprietary |
| Context window | 1M | 1.05M |
| Max output | 393K | 66K |
| Input $ / M tokens | $0.66 | $0.30 |
| Output $ / M tokens | $1.98 | $2.50 |
| Results tracked | 48 | 54 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
DeepSeek V4 Pro: 52.4 (#34), Gemini 2.5 Flash: 35.8 (#220)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| WeirdML | 66.2% | 41.9% |
| LMArena Coding | 1470 | 1424 |
| ALE-Bench | 1,403 | 661.88 |
| SWE-bench Verified | 77.6% | — |
| FrontierCode | 28.6% | — |
| SWE-bench Verified (bash only) | — | 28.7% |
| Aider Polyglot | — | 55.1% |
| LMArena WebDev | 1582 | — |
| SciCode | 51% | — |
Agentic & Tool Use DeepSeek V4 Pro leads
DeepSeek V4 Pro: 32.8 (#58), Gemini 2.5 Flash: 30.8 (#74)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| Vending-Bench 2 | 3,285 | 548.84 |
| Terminal-Bench | — | 17.1% |
| APEX-Agents | 47.3% | — |
| Berkeley Function Calling Leaderboard | — | 56.2% |
| TheAgentCompany | — | 41.1% |
| BALROG | — | 33.5% |
Reasoning DeepSeek V4 Pro leads
DeepSeek V4 Pro: 56.5 (#24), Gemini 2.5 Flash: 18.1 (#286)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| ARC-AGI-2 | 61.3% | 2.5% |
| Kagi LLM Benchmark | 53.5% | 56.8% |
| ARC-AGI-1 | 90.5% | 33.3% |
| CritPt | 18% | 1.1% |
| LMArena Hard Prompts | 1461 | 1422 |
| DTBench | 93.9% | 76.5% |
| LMCA | 45.5% | 27.5% |
| Epoch Capabilities Index | 155.31 | 143.03 |
| ForecastBench | 56.1 | 60.6 |
| SimpleBench | — | 41.2% |
| NYT Connections (extended) | 91.3% | — |
| Chess Puzzles | 47% | — |
| EnigmaEval | — | 2.7% |
| Mystery Game Puzzles | 43% | — |
| Surface Evolver Bench | 40% | — |
Math DeepSeek V4 Pro leads
DeepSeek V4 Pro: 64.8 (#30), Gemini 2.5 Flash: 39.9 (#98)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.6% | 73.1% |
| LMArena Math | 1455 | 1415 |
| FrontierMath (Tiers 1-3) | 64.6% | — |
| FrontierMath Tier 4 | 26.8% | — |
| MathArena Final-Answer Competitions | 76.6% | — |
| ProofBench | 50% | — |
| Omni-MATH | — | 38.5% |
| FrontierMath (Feb 2025 set) | — | 4.8% |
| FrontierMath Tier 4 (v1) | — | 4.2% |
Knowledge DeepSeek V4 Pro leads
DeepSeek V4 Pro: 59.5 (#31), Gemini 2.5 Flash: 36.4 (#168)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| Vectara Hallucination Rate | 8.6% | 7.8% |
| LMArena Expert | 1464 | 1426 |
| GPQA Diamond | 91.7% | — |
| Humanity's Last Exam | — | 12.1% |
| SimpleQA Verified | 52.9% | — |
| MMLU-Pro | — | 63.9% |
| Confabulations | — | 16.8% |
| GPQA (HELM) | — | 39% |
Multimodal Not comparable
DeepSeek V4 Pro: —, Gemini 2.5 Flash: 41.8 (#32)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| LMArena Vision | — | 1253 |
| GeoBench | — | 76% |
| VPCT | — | 46.2% |
| SpatialViz-Bench | — | 36.9% |
Multilingual DeepSeek V4 Pro leads
DeepSeek V4 Pro: 54.4 (#45), Gemini 2.5 Flash: 52.3 (#88)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| LMArena Non-English | 1439 | 1409 |
| LMArena Chinese | 1486 | 1450 |
| LMArena French | 1472 | 1433 |
| LMArena German | 1458 | 1418 |
| LMArena Japanese | 1445 | 1405 |
| LMArena Korean | 1447 | 1385 |
| LMArena Russian | 1453 | 1415 |
| LMArena Spanish | 1458 | 1421 |
Instruction Following Too close to call
DeepSeek V4 Pro: 76.1 (#47), Gemini 2.5 Flash: 75.7 (#54)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| LMArena Instruction Following | 1448 | 1405 |
| IFEval | — | 89.8% |
Long Context Gemini 2.5 Flash leads
DeepSeek V4 Pro: 45.0 (#51), Gemini 2.5 Flash: 47.5 (#17)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| LMArena Longer Query | 1458 | 1419 |
| Fiction.LiveBench | — | 77.8% |
| CL-bench Life | 13.5% | — |
Writing & Preference DeepSeek V4 Pro leads
DeepSeek V4 Pro: 65.5 (#46), Gemini 2.5 Flash: 53.8 (#157)
| Benchmark | DeepSeek V4 Pro | Gemini 2.5 Flash |
|---|---|---|
| LMArena Text | 1451 | 1417 |
| LMArena Creative Writing | 1446 | 1400 |
| EQ-Bench Creative Writing | 1553 | 1137 |
| LMArena Multi-Turn | 1467 | 1408 |
| Short-Story Creative Writing | — | 76.5% |
| WildBench | — | 81.7% |
| EQ-Bench 4 | 1166 | — |
Frequently asked questions
Is DeepSeek V4 Pro better than Gemini 2.5 Flash?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 39.3 on the Noometry Index.
Which is cheaper, DeepSeek V4 Pro or Gemini 2.5 Flash?
Gemini 2.5 Flash is cheaper. It lists at $0.30 per million input tokens and $2.50 per million output tokens; DeepSeek V4 Pro lists at $0.66 and $1.98.
Is DeepSeek V4 Pro or Gemini 2.5 Flash better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 35.8 in the Noometry coding category.
Which has the bigger context window?
Gemini 2.5 Flash does, with 1.05M tokens against 1M.
How many benchmarks do DeepSeek V4 Pro and Gemini 2.5 Flash share?
31 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and Gemini 2.5 Flash has 54.