DeepSeek V4 Flash vs Gemini 3 Pro
Full profiles: DeepSeek V4 Flash (DeepSeek) and Gemini 3 Pro (Google): every benchmark, price and rank.
Updated · Data from 15 sources
As of 16 Nov 2026, DeepSeek V4 Flash and Gemini 3 Pro are neck and neck on quality (57.2 vs 60.0 on the BenchLeader Index, inside the ±5.2 margin), and DeepSeek V4 Flash is 17× cheaper at $0.262 vs $4.50 per 1M tokens blended. Choose DeepSeek V4 Flash for maths (62 vs 54); choose Gemini 3 Pro for human preference (68 vs 62), reasoning (63 vs 59), and coding (58 vs 55).
Key facts (as of 16 Nov 2026)
- DeepSeek V4 Flash scores 57.2 on the BenchLeader Index and Gemini 3 Pro scores 60.0 (BenchLeader, 16 Nov 2026).
- DeepSeek V4 Flash costs $0.262 per 1M tokens blended (list price, 16 Nov 2026).
- Gemini 3 Pro costs $4.50 per 1M tokens blended (list price, 16 Nov 2026).
- Of the 15 benchmarks both models have, Gemini 3 Pro leads 11 (BenchLeader, 16 Nov 2026).
- On Arena WebDev, DeepSeek V4 Flash is rated 1582 and Gemini 3 Pro 1441 (LMArena, as published by 16 Nov 2026).
Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Coding | Gemini 3 Pro | 58 vs 55 category score; WeirdML 69.9% vs 57.0% | Medium |
| Maths | DeepSeek V4 Flash | 62 vs 54 category score; Arena Maths 1441 vs 1475 | High |
| Reasoning | Gemini 3 Pro | 63 vs 59 category score; Arena Hard Prompts 1502 vs 1459 | Medium |
| Knowledge | Gemini 3 Pro | 56 vs 55 category score; MMLU-Pro 90.1% vs 86.2% | Medium |
| Chat quality (human preference) | Gemini 3 Pro | 68 vs 62 on Arena votes | Medium |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- DeepSeek V4 Flash (high) and Gemini 3 Pro are neck and neck on quality (57.2 vs 60.0, within the ±5.2 margin).
- DeepSeek V4 Flash (high) is stronger in maths.
- Gemini 3 Pro is stronger in coding, human preference, knowledge, reasoning.
- DeepSeek V4 Flash (high) is 17× cheaper ($0.262 vs $4.50 per 1M blended).
At a glance
| Spec | DeepSeek V4 Flash | Gemini 3 Pro |
|---|---|---|
| Developer | DeepSeek | |
| Released | 31 Jul 2026 | 18 Nov 2025 |
| Context window (tokens) | 1,048,576 | – |
| Compared at | DeepSeek V4 Flash (high) | Gemini 3 Pro |
| Open weights | Yes | No |
| Price in / out / cached input per 1M | $0.150 / $0.600 / $0.0030 | – |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | DeepSeek V4 Flash (high) | Gemini 3 Pro | Leader |
|---|---|---|---|
| Reasoning | 59.2 | 62.6 | Gemini 3 Pro |
| Coding | 55.0 | 57.8 | Gemini 3 Pro |
| Agents & tools | – | 61.7 | not comparable |
| Maths | 61.9 | 54.0 | DeepSeek V4 Flash |
| Knowledge | 54.8 | 55.9 | Gemini 3 Pro |
| Instruction following | – | 61.1 | not comparable |
| Human preference | 62.0 | 67.7 | Gemini 3 Pro |
| Multimodal | – | 65.6 | not comparable |
On one chart
Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.
Not drawn, measured for fewer than two of these models: Agents & tools, Instruction following, Long context, Multimodal.
Where sources disagree
Across the 15 benchmarks both have, DeepSeek V4 Flash leads 4, Gemini 3 Pro leads 11. Every overall index that rates both puts Gemini 3 Pro ahead.
| Overall index | DeepSeek V4 Flash (high) | Gemini 3 Pro | Leader |
|---|---|---|---|
| BenchLeader Index | 57.2 | 60.0 | neck and neck |
| Arena Text | 1439 | 1486 | Gemini 3 Pro |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.
| Measure | DeepSeek V4 Flash (high) | Gemini 3 Pro |
|---|---|---|
| Blended price per 1M tokens | $0.262 | – |
| Input / output per 1M | $0.150 / $0.600 | – / – |
| Cached input per 1M | $0.0030 | – |
| Price per task: chat reply | $0.0002 | – |
| Price per task: summarize a 30-page report | $0.0022 | – |
| Price per task: code edit | $0.0018 | – |
| Price per task: agentic coding session | $0.011 | – |
| Price per task: structured extraction | $0.0004 | – |
| Agentic coding session, input mostly cached | $0.0048 | – |
| Output speed | 82 tok/s | – |
| Time to first answer | 0.7 s | – |
Each model is compared at its best-ranked reasoning-effort setting. DeepSeek V4 Flash (high) scores 57.2 and takes 1 s to its first answer token; DeepSeek V4 Flash (max) scores 56.8 in 28 s, DeepSeek V4 Flash scores 53.8 in 1 s. Gemini 3 Pro scores 60.0; Gemini 3 Pro (high) scores 60.0. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
147 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | DeepSeek V4 Flash (high) | Gemini 3 Pro |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 57.2 | 60.0 |
| Coding score | BenchLeader | 55.0 | 57.8 |
| Human preference score | BenchLeader | 62.0 | 67.7 |
| Knowledge score | BenchLeader | 54.8 | 55.9 |
| Maths score | BenchLeader | 61.9 | 54.0 |
| Reasoning score | BenchLeader | 59.2 | 62.6 |
| Agents & tools score | BenchLeader | – | 61.7 |
| Instruction following score | BenchLeader | – | 61.1 |
| Multimodal score | BenchLeader | – | 65.6 |
| Blended price $/M | BenchLeader | $0.262 | – |
| Output speed | BenchLeader | 82 tok/s | – |
| Time to first answer | BenchLeader | 0.7 s | – |
| Context window | BenchLeader | 1,048,576 tokens | – |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 91.0%† | 92.6% |
| Humanity's Last Exam | Scale AI / CAIS | – | 37.5% |
| SimpleBench | SimpleBench | 61.1%† | 76.4% |
| Arena Hard Prompts | Arena (LMArena) | 1459 | 1502 |
| LiveBench 1.0 Reasoning | LiveBench | 86.6%† | – |
| GPQA Diamond (AA) | Artificial Analysis | 90.8%† | 90.8%† |
| Humanity's Last Exam (AA) | Artificial Analysis | 38.5%† | 39.7%† |
| GPQA Diamond (Vals) | Vals AI | 89.9% | 91.7%† |
| Kagi LLM Benchmark | Kagi LLM Benchmark | 52.2%† | 80.1% |
| GPQA Diamond (HELM) | HELM Capabilities | – | 80.3% |
| ARC-AGI-1 | ARC Prize | 87.0% | 75.0% |
| ARC-AGI-2 | ARC Prize | 56.0% | 31.1% |
| CritPt | Artificial Analysis | 16.6%† | 9.1%† |
| SciPredict | Scale AI SEAL | – | 25.3% |
| Arena Expert | Arena (LMArena) | 1467 | 1499 |
| Chess Puzzles | Epoch AI Benchmarking Hub | 32.0%† | – |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | 34.0%† | – |
| CL-bench | Epoch AI Benchmarking Hub | – | 15.8% |
| LMCA | Epoch AI Benchmarking Hub | 41.7%† | – |
| DTBench | Epoch AI Benchmarking Hub | 90.9%† | – |
| ForecastBench | ForecastBench | – | 65.0% |
| Coding | |||
| SWE-bench Verified (Epoch) | Epoch AI Benchmarking Hub | – | 72.9% |
| SciCode | SciCode | 42.0% | – |
| WeirdML | WeirdML | 57.0% | 69.9% |
| FrontierCode | Cognition | 18.8%† | – |
Show all 147 benchmarks
| Metric | Publisher | DeepSeek V4 Flash (high) | Gemini 3 Pro |
|---|---|---|---|
| Coding | |||
| GSO-Bench | GSO-Bench | – | 18.6% |
| Arena Coding | Arena (LMArena) | 1480 | 1517 |
| Arena WebDev | Arena (LMArena) | 1582 | 1441 |
| LiveBench 1.0 Coding | LiveBench | 75.0%† | – |
| SciCode (AA) | Artificial Analysis | 50.4%† | – |
| LiveCodeBench | Vals AI | 87.3% | 86.4% |
| IOI | Vals AI | 32.7% | – |
| SWE-bench (Vals) | Vals AI | 88.8% | 76.4%† |
| SWE-Bench Pro | Scale AI SEAL | – | 43.3% |
| SWE-bench Verified (bash only) | SWE-bench | – | 74.2% |
| SWE-bench Multilingual (bash only) | SWE-bench | – | 68.7% |
| SWE-bench Verified (any scaffold) | SWE-bench | – | 77.4% |
| Code Migration | Vals AI | 38.6% | – |
| ProgramBench | Vals AI | 0.0% | – |
| Vibe Code Bench v1.1 | Vals AI | 74.7% | 14.3%† |
| SWE-Bench Pro (private) | Scale AI SEAL | – | 17.9% |
| ALE-Bench | Epoch AI Benchmarking Hub | 678.2 | 1176.8 |
| AlgoTune | AlgoTune | – | 1.8 |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | 12.1%† | – |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | 78.7%† | – |
| GameBench 2 Overall | DuelLab GameBench 2 | 34.1† | – |
| GameBench 2 by reasoning setting | DuelLab GameBench 2 | 43.4 | – |
| Design Arena Website | Design Arena | 1199† | 1244 |
| LiveCodeBench (AA) | Artificial Analysis | – | 91.8%† |
| Agents & tools | |||
| Terminal-Bench | Terminal-Bench | – | 69.4% |
| GDPval | OpenAI | – | 40.3% |
| Remote Labor Index | Scale AI / CAIS | – | 1.3% |
| LiveBench Agentic Coding | LiveBench | 46.8%† | – |
| Terminal-Bench Hard | Artificial Analysis | – | 41.7%† |
| τ²-Bench Telecom (AA) | Artificial Analysis | – | 87.1%† |
| Terminal-Bench 2.1 (Vals) | Vals AI | 67.0%† | – |
| MCP Atlas | Scale AI SEAL | – | 70.3% |
| HiL-Bench | Scale AI SEAL | 47.8%† | – |
| BFCL Overall | Berkeley Function Calling Leaderboard | – | 72.5% |
| τ²-bench | τ²-bench | – | 82.5%† |
| τ³-bench Banking | τ²-bench | – | 18.0%† |
| GDPval-AA v2.1 | Artificial Analysis | 1437† | – |
| τ³-Banking (AA) | Artificial Analysis | 39.4%† | – |
| CyberBench | Vals AI | 72.1% | – |
| Finance Agent v2 | Vals AI | 49.5% | – |
| Harvey's Legal Agent Benchmark v1 | Vals AI | 8.3% | – |
| Legal Research Bench | Vals AI | 30.3% | – |
| Poker Agent | Vals AI | – | 1078.9% |
| SkillsBench | Vals AI | 50.7% | – |
| Tax Agent Bench | Vals AI | 28.1% | – |
| Terminal-Bench 2.0 (Vals) | Vals AI | – | 55.1%† |
| Terminal-Bench 4.0 (Vals) | Vals AI | 18.7% | – |
| Terminal-Bench Science | Vals AI | 4.3% | – |
| Arena Search | Arena (LMArena) | – | 1201 |
| BALROG | BALROG | – | 58.1% |
| DeepResearch Bench | DeepResearch Bench | – | 46.3%† |
| METR Time Horizons | METR | – | 71.0% |
| Vending-Bench 2 | Andon Labs | – | 5478.2 |
| AutomationBench | Artificial Analysis | 54.0%† | – |
| GDP.pdf (AA) | Artificial Analysis | 11.0%† | – |
| AA-Briefcase v1.1 | Artificial Analysis | 1255† | – |
| Design Arena Agentic | Design Arena | – | 1080 |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 57.5%† | – |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 24.4%† | – |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | 94.4%† | 91.4% |
| ProofBench | Vals AI | 56.0%† | 20.0% |
| LiveBench Mathematics | LiveBench | 86.8%† | – |
| AIME (Vals) | Vals AI | – | 96.7%† |
| MGSM | Vals AI | – | 93.9%† |
| AIME 2026 | MathArena | 95.8%† | 91.7% |
| HMMT February 2026 | MathArena | 93.9%† | 86.4% |
| MathArena Apex | MathArena | 27.1%† | 23.4% |
| Omni-MATH (HELM) | HELM Capabilities | – | 55.6% |
| MATH 500 | Vals AI | – | 96.4%† |
| Arena Maths | Arena (LMArena) | 1441 | 1475 |
| AIME 2025 (AA) | Artificial Analysis | – | 95.7%† |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 33.6%† | – |
| LiveBench 1.0 Data Analysis | LiveBench | 79.3%† | – |
| MMLU-Pro | Vals AI | 86.2% | 90.1% |
| LegalBench | Vals AI | 77.7% | 87.0%† |
| CorpFin | Vals AI | 61.9% | 63.7%† |
| TaxEval | Vals AI | 70.7% | 72.6%† |
| MedQA | Vals AI | – | 96.0%† |
| PRBench Finance | Scale AI SEAL | – | 39.2% |
| PRBench Legal | Scale AI SEAL | – | 40.6% |
| MultiNRC | Scale AI SEAL | – | 59.0% |
| MMLU-Pro (HELM) | HELM Capabilities | – | 90.3% |
| BioMysteryBench | Vals AI | 64.4% | – |
| CaseLaw v2 | Vals AI | – | 53.1%† |
| Excel Modeling Benchmark | Vals AI | 57.0% | – |
| MedCode | Vals AI | 41.4% | 52.2%† |
| MedScribe | Vals AI | 80.4% | 72.0%† |
| Global-MMLU-Lite (AA) | Artificial Analysis | – | 92.2%† |
| Instruction following | |||
| LiveBench Language | LiveBench | 79.2%† | – |
| LiveBench Instruction Following | LiveBench | 65.5%† | – |
| IFBench | Artificial Analysis | – | 70.4%† |
| MultiChallenge | Scale AI SEAL | – | 65.7% |
| TutorBench | Scale AI SEAL | – | 53.7% |
| IFEval (HELM) | HELM Capabilities | – | 87.6% |
| WildBench (HELM) | HELM Capabilities | – | 85.9% |
| Arena Instruction Following | Arena (LMArena) | 1434 | 1473 |
| Human preference | |||
| Arena Text | Arena (LMArena) | 1439 | 1486 |
| EQ-Bench Creative Writing v3 | EQ-Bench | 1559† | 1524 |
| Arena Creative Writing | Arena (LMArena) | 1411 | 1484 |
| Arena Multi-turn | Arena (LMArena) | 1447 | 1495 |
| Arena Longer Queries | Arena (LMArena) | 1449 | 1491 |
| Multimodal | |||
| Arena Vision | Arena (LMArena) | – | 1305 |
| MMMU-Pro | Artificial Analysis | – | 80.2%† |
| VISTA | Scale AI SEAL | – | 51.5% |
| MMMU-Pro (Vals) | Vals AI | – | 87.5%† |
| MortgageTax | Vals AI | – | 69.1%† |
| AudioMC | Scale AI SEAL | – | 54.6%† |
| AudioMC (text output) | Scale AI SEAL | – | 54.6%† |
| VTB | Scale AI SEAL | – | 26.9% |
| GeoBench | GeoBench | – | 84.0% |
| VPCT | Epoch AI Benchmarking Hub | – | 91.0% |
| Long context | |||
| AA-LCR v1.1 | Artificial Analysis | 79.7%† | 76.0%† |
| Arena Document | Arena (LMArena) | – | 1451 |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 154.4† | 153.0 |
| LiveBench 1.0 | LiveBench | 74.2%† | – |
| AA Intelligence Index v4.3.2 | Artificial Analysis | 34.3† | 28.0† |
| AA Progress Index v1 | Artificial Analysis | 309.4† | 290.7† |
| HELM Capabilities mean | HELM Capabilities | – | 79.9% |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | – | 41.7% |
| MASK | Scale AI SEAL | – | 42.6% |
| PropensityBench | Scale AI SEAL | – | 52.9% |
| Metadata | |||
| AA Openness Index | Artificial Analysis | 44.4† | 5.6† |
Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.
DuelLab GameBench 2, release gb2-20261111-7b598bb13de4 (https://benchmarks.duellab.org/releases/gb2-20261111-7b598bb13de4/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.
What changed
- Gemini 3 Pro: 1 score revision.
- DeepSeek V4 Flash (high): 1 score revision.
- Gemini 3 Pro: 1 new result (Arena Search).
- DeepSeek V4 Flash (high): 1 score revision.
- DeepSeek V4 Flash (high): 1 new result (GameBench 2 by reasoning setting).
- DeepSeek V4 Flash (high): 1 new result (Harvey's Legal Agent Benchmark v1).
- Gemini 3 Pro: 1 score revision.
- DeepSeek V4 Flash (high): BenchLeader Index 59.3 → 57.3.
- DeepSeek V4 Flash (high): input price moved from $0.11/M to $0.15/M (36% up); output price moved from $0.23/M to $0.60/M (161% up).
- Gemini 3 Pro: 1 score revision.
Related comparisons
- DeepSeek V4 Flash vs Claude Fable 5.1
- DeepSeek V4 Flash vs GPT-6.1 Sol
- Gemini 3 Pro vs Gemini 4 Carbon
- Gemini 3 Pro vs Claude Opus 5.5
- Gemini 3 Pro vs GPT-6 Astra
- DeepSeek V4 Flash vs Gemini 4 Argon
- DeepSeek V4 Flash vs Claude Sonnet 5.5
- DeepSeek V4 Flash vs Muse Spark 1.3
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
DeepSeek V4 Flash vs Gemini 3 Pro: questions
- Is DeepSeek V4 Flash better than Gemini 3 Pro?
- DeepSeek V4 Flash (high) and Gemini 3 Pro are neck and neck on quality (57.2 vs 60.0, within the ±5.2 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 16 Nov 2026, so check the category scores for your use.
- DeepSeek V4 Flash vs Gemini 3 Pro: which is cheaper, and will that change?
- DeepSeek V4 Flash is cheaper today: $0.262 against $4.50 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Is DeepSeek V4 Flash or Gemini 3 Pro better for coding?
- Gemini 3 Pro scores higher in coding (58 vs 55 on the category index, where 50 is average). On Arena WebDev, the top shared coding benchmark, it is 1582 for DeepSeek V4 Flash against 1441 for Gemini 3 Pro.
- Do independent benchmarks agree on DeepSeek V4 Flash vs Gemini 3 Pro?
- Across the 15 benchmarks both have, DeepSeek V4 Flash leads 4, Gemini 3 Pro leads 11. Every overall index that rates both puts Gemini 3 Pro ahead.
- When was DeepSeek V4 Flash released?
- DeepSeek V4 Flash was released on 31 Jul 2026.