DeepSeek V4.1 Flash 1017 vs GPT-5
DeepSeek V4.1 Flash 1017 leads GPT-5 on coding, 70.3 to 38.8 on the Vibe Coding Index, and DeepSeek V4.1 Flash 1017 costs 16.7x less per output token ($0.60 vs $10.00 per million). Updated October 22, 2026; 21 benchmarks have a result for both.
DeepSeek V4.1 Flash 1017 or GPT-5: which to pick
Pick DeepSeek V4.1 Flash 1017 for coding: it scores 70.3 on the Vibe Coding Index against 38.8 for GPT-5.
Pick GPT-5 for work measured by SimpleQA Verified, where it leads.
- Shared results
- 21
- DeepSeek V4.1 Flash 1017 only
- 18
- GPT-5 only
- 92
- Like-for-like categories
- 7 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
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, six public coding benchmarks
DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 scores 70.3 against 38.8.
4 and 3 benchmarks behind the scores
Intelligence
AA Intelligence Index
DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 leads, 39.5 against 23.0.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only GPT-5 has a GPQA Diamond result, 85.6%.
Directional only
Math
AIME
No call
Only GPT-5 has a AIME result, 93.4%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5 has a MMLU Pro result, 86.5%.
Directional only
Web design
Design Arena: websites
DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 leads, 1311 against 1184.
Same board, both measured
General
Vals Index
No call
Only DeepSeek V4.1 Flash 1017 has a Vals Index result, 61.3%.
Directional only
Long documents
Listed context window
DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 takes 1.05M tokens against 400K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V4.1 Flash 1017
DeepSeek V4.1 Flash 1017 $0.0004, GPT-5 $0.0063 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5 has the number, $0.39.
Directional only
At a glance
- Coding index
- DeepSeek V4.1 Flash 1017 70.3 from 4 benchmarks, GPT-5 38.8 from 3 benchmarks.
- Widest category
- Coding, 39.7 points apart on average over 4 shared benchmarks.
- Widest single gap
- Vibe Code Bench (Vals AI): DeepSeek V4.1 Flash 1017 87.7% against GPT-5 20.1%, 67.6 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V4.1 Flash 1017 $0.60 against GPT-5 $10.00 per million output tokens, 16.7x apart.
- Context window
- DeepSeek V4.1 Flash 1017 takes 1.05M tokens, GPT-5 400K.
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 | DeepSeek V4.1 Flash 1017 | Bars | GPT-5 | Ties |
|---|---|---|---|---|
| Coding | 4 | 0 | ||
| Intelligence | 1 | 0 | ||
| Reasoning | 4 | 0 | ||
| Math | 3 | 0 | ||
| Knowledge | 0 | 1 | ||
| Web design | 7 | 0 | ||
| General | 1 | 0 | ||
| All shared boards | 20 | 1 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index40
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| Aider polyglot | Not measured | 88.0% | Only GPT-5 measured |
| SWE-bench Verified | Not measured | 74.4% | Only GPT-5 measured |
| LiveBench coding | 96.0% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LMArena WebDev | 1664 | 1418 | DeepSeek V4.1 Flash 1017 |
| AA Coding Index | Not measured | 37.8 | Only GPT-5 measured |
| LiveCodeBench (Vals AI) | Not measured | 85.9% | Only GPT-5 measured |
| SWE-bench Verified (Vals AI) | Not measured | 69.0% | Only GPT-5 measured |
| Terminal-Bench 4 (Vals AI) | 31.8% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Vibe Code Bench (Vals AI) | 87.7% | 20.1% | DeepSeek V4.1 Flash 1017 |
| Code migration (Vals AI) | 44.2% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 84.6% | Only GPT-5 measured |
| SciCode | 55.9% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 37.9% | Only GPT-5 measured |
| Terminal-Bench | Not measured | 49.6% | Only GPT-5 measured |
| GSO-Bench | Not measured | 6.9% | Only GPT-5 measured |
| WeirdML | Not measured | 60.7% | Only GPT-5 measured |
| ALE-Bench | 1622.0 | 807.6 | DeepSeek V4.1 Flash 1017 |
| LMArena Text: coding | Not measured | 1436 | Only GPT-5 measured |
| LMArena Text: software and IT | Not measured | 1420 | Only GPT-5 measured |
| LMArena WebDev: React | 1667 | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LMArena WebDev: HTML | 1687 | 1427 | DeepSeek V4.1 Flash 1017 |
| SciCode | Not measured | 0.0% | Only GPT-5 measured |
| Neither model has a result on the other 17 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| AA Intelligence Index | 39.5 | 23.0 | DeepSeek V4.1 Flash 1017 |
| Epoch Capabilities Index | Not measured | 150.0 | Only GPT-5 measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index16 boards, none shared
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| Design Arena: spreadsheets | 1245 | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LiveBench research | 79.5% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LiveBench browser | 76.0% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LiveBench long agentic | 50.9% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Finance Agent (Vals AI) | 57.0% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Tax Agent Bench (Vals AI) | 44.4% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Legal Research Bench (Vals AI) | 45.2% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Harvey's Legal Agent Benchmark (Vals AI) | 14.2% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| Excel Modeling Benchmark (Vals AI) | 61.8% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| τ²-Bench Telecom | Not measured | 86.5% | Only GPT-5 measured |
| METR time horizons | Not measured | 69.6% | Only GPT-5 measured |
| Remote Labor Index | Not measured | 1.7% | Only GPT-5 measured |
| GDPval | Not measured | 34.8% | Only GPT-5 measured |
| BALROG | Not measured | 32.8% | Only GPT-5 measured |
| DeepResearch Bench | Not measured | 49.6% | Only GPT-5 measured |
| BrowseComp-Plus | Not measured | 80.0% | Only GPT-5 measured |
| Neither model has a result on the other 25 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 85.6% | Only GPT-5 measured |
| LiveBench reasoning | 89.1% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| MMMU Pro (Vals AI) | Not measured | 81.5% | Only GPT-5 measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 85.4% | Only GPT-5 measured |
| Humanity's Last Exam (Artificial Analysis run) | 43.3% | 28.5% | DeepSeek V4.1 Flash 1017 |
| GPQA Diamond (Epoch AI run) | 93.2% | 86.2% | DeepSeek V4.1 Flash 1017 |
| ARC-AGI | Not measured | 65.7% | Only GPT-5 measured |
| ARC-AGI-2 | Not measured | 9.9% | Only GPT-5 measured |
| SimpleBench | Not measured | 56.7% | Only GPT-5 measured |
| Chess puzzles | 46.0% | 36.0% | DeepSeek V4.1 Flash 1017 |
| Mystery game puzzles | 49.0% | 23.0% | DeepSeek V4.1 Flash 1017 |
| EBR-bench | Not measured | 12.7% | Only GPT-5 measured |
| VPCT | Not measured | 66.0% | Only GPT-5 measured |
| LMArena Text: hard prompts | Not measured | 1415 | Only GPT-5 measured |
| LMArena Text: expert | Not measured | 1419 | Only GPT-5 measured |
| LMArena Text: hard prompts in English | Not measured | 1419 | Only GPT-5 measured |
| LMArena Vision: diagrams | Not measured | 1252 | Only GPT-5 measured |
| LMArena Vision: homework | Not measured | 1278 | Only GPT-5 measured |
| Kagi LLM Benchmark | Not measured | 72.7% | Only GPT-5 measured |
| Neither model has a result on the other 3 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| AIME | Not measured | 93.4% | Only GPT-5 measured |
| MATH 500 (Vals AI) | Not measured | 96.0% | Only GPT-5 measured |
| MGSM (Vals AI) | Not measured | 92.8% | Only GPT-5 measured |
| AA Math Index | Not measured | 94.3 | Only GPT-5 measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 94.3% | Only GPT-5 measured |
| OTIS Mock AIME | 99.4% | 91.4% | DeepSeek V4.1 Flash 1017 |
| FrontierMath, tiers 1 to 3 | 71.2% | 55.4% | DeepSeek V4.1 Flash 1017 |
| FrontierMath, tier 4 | 46.3% | 22.0% | DeepSeek V4.1 Flash 1017 |
| ProofBench | Not measured | 18.0% | Only GPT-5 measured |
| MATH level 5 | Not measured | 98.1% | Only GPT-5 measured |
| LMArena Text: math | Not measured | 1407 | Only GPT-5 measured |
| LMArena Text: mathematical work | Not measured | 1405 | Only GPT-5 measured |
| Neither model has a result on the other 1 board here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 86.5% | Only GPT-5 measured |
| LegalBench (Vals AI) | Not measured | 86.0% | Only GPT-5 measured |
| CaseLaw (Vals AI) | Not measured | 66.5% | Only GPT-5 measured |
| CorpFin (Vals AI) | Not measured | 61.1% | Only GPT-5 measured |
| TaxEval (Vals AI) | Not measured | 73.4% | Only GPT-5 measured |
| MortgageTax (Vals AI) | Not measured | 65.5% | Only GPT-5 measured |
| MedQA (Vals AI) | Not measured | 96.3% | Only GPT-5 measured |
| MedCode (Vals AI) | Not measured | 49.6% | Only GPT-5 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 87.1% | Only GPT-5 measured |
| SimpleQA Verified | 42.5% | 50.1% | GPT-5 |
| GeoBench | Not measured | 81.0% | Only GPT-5 measured |
| LMArena Text: science | Not measured | 1416 | Only GPT-5 measured |
| LMArena Text: medicine and healthcare | Not measured | 1413 | Only GPT-5 measured |
| LMArena Text: legal and government | Not measured | 1431 | Only GPT-5 measured |
| LMArena Text: business and finance | Not measured | 1400 | Only GPT-5 measured |
| FACTS Benchmark Suite | Not measured | 56.5% | Only GPT-5 measured |
| Neither model has a result on the other 2 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index70
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| Design Arena: websites | 1311 | 1184 | DeepSeek V4.1 Flash 1017 |
| Design Arena: UI components | 1303 | 1177 | DeepSeek V4.1 Flash 1017 |
| Design Arena: game dev | 1357 | 1185 | DeepSeek V4.1 Flash 1017 |
| Design Arena: data viz | 1330 | 1221 | DeepSeek V4.1 Flash 1017 |
| Design Arena: 3D | 1358 | 1055 | DeepSeek V4.1 Flash 1017 |
| Design Arena: all code categories | 1332 | 1167 | DeepSeek V4.1 Flash 1017 |
| Design Arena: SVG | Not measured | 1188 | Only GPT-5 measured |
| Design Arena: ASCII art | 1242 | 1141 | DeepSeek V4.1 Flash 1017 |
| LiveBench design | 83.4% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | Leader |
|---|---|---|---|
| Vals Index | 61.3% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| LiveBench data analysis | 79.5% | Not measured | Only DeepSeek V4.1 Flash 1017 measured |
| SAGE (Vals AI) | Not measured | 43.7% | Only GPT-5 measured |
| MedScribe (Vals AI) | Not measured | 83.7% | Only GPT-5 measured |
| IFBench | Not measured | 73.1% | Only GPT-5 measured |
| Long context reasoning (AA-LCR) | 82.7% | 78.2% | DeepSeek V4.1 Flash 1017 |
| DTBench | Not measured | 90.7% | Only GPT-5 measured |
| LMCA | Not measured | 40.0% | Only GPT-5 measured |
| Fiction.LiveBench | Not measured | 97.2% | Only GPT-5 measured |
| Creative writing (Lech Mazur) | Not measured | 86.0% | Only GPT-5 measured |
| LMArena Text | Not measured | 1406 | Only GPT-5 measured |
| LMArena Text: instruction following | Not measured | 1388 | Only GPT-5 measured |
| LMArena Text: creative writing | Not measured | 1365 | Only GPT-5 measured |
| LMArena Vision | Not measured | 1232 | Only GPT-5 measured |
| LMArena Text: English | Not measured | 1409 | Only GPT-5 measured |
| LMArena Text: non-English | Not measured | 1397 | Only GPT-5 measured |
| LMArena Text: Chinese | Not measured | 1422 | Only GPT-5 measured |
| LMArena Text: German | Not measured | 1416 | Only GPT-5 measured |
| LMArena Text: Japanese | Not measured | 1409 | Only GPT-5 measured |
| LMArena Text: Korean | Not measured | 1360 | Only GPT-5 measured |
| LMArena Text: French | Not measured | 1409 | Only GPT-5 measured |
| LMArena Text: Spanish | Not measured | 1399 | Only GPT-5 measured |
| LMArena Text: Russian | Not measured | 1407 | Only GPT-5 measured |
| LMArena Text: Polish | Not measured | 1431 | Only GPT-5 measured |
| LMArena Text: multi-turn | Not measured | 1426 | Only GPT-5 measured |
| LMArena Text: longer queries | Not measured | 1399 | Only GPT-5 measured |
| LMArena Text: writing and language | Not measured | 1384 | Only GPT-5 measured |
| LMArena Text: entertainment and media | Not measured | 1379 | Only GPT-5 measured |
| LMArena Vision: English | Not measured | 1231 | Only GPT-5 measured |
| LMArena Vision: Chinese | Not measured | 1267 | Only GPT-5 measured |
| LMArena Vision: reading text | Not measured | 1246 | Only GPT-5 measured |
| LMArena Vision: humor | Not measured | 1219 | Only GPT-5 measured |
| LMArena Vision: captioning | Not measured | 1212 | Only GPT-5 measured |
| LMArena Vision: recognizing things | Not measured | 1263 | Only GPT-5 measured |
| LMArena Vision: creative writing | Not measured | 1210 | Only GPT-5 measured |
| Neither model has a result on the other 11 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V4.1 Flash 1017 | GPT-5 | 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
DeepSeek V4.1 Flash 1017 costs less
Repository review
50K input + 3K output tokens
DeepSeek V4.1 Flash 1017 costs less
Agent loop
200K input + 10K output tokens
DeepSeek V4.1 Flash 1017 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 | DeepSeek V4.1 Flash 1017 | OpenAI GPT-5 |
|---|---|---|
| Maker | DeepSeek | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | October 17, 2026 | August 7, 2025 |
| Context window | 1.05M | 400K |
| Input $/Mtok | $0.15 | $1.25 |
| Output $/Mtok | $0.60 | $10.00 |
| Coding index | 70.3 | 38.8 |
| Independent measurements behind it | 4 | 3 |
| Cost per solved task | Not computable | $0.39 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V4.1 Flash 1017 better than GPT-5 for coding?
Pick DeepSeek V4.1 Flash 1017 for coding: it scores 70.3 on the Vibe Coding Index against 38.8 for GPT-5. Pick GPT-5 for work measured by SimpleQA Verified, where it leads.
Which is cheaper, DeepSeek V4.1 Flash 1017 or GPT-5?
DeepSeek V4.1 Flash 1017 costs $0.15 per million input tokens and $0.60 per million output tokens; GPT-5 costs $1.25 and $10.00. Cost per solved task is known for GPT-5 only.
Which has the bigger context window, DeepSeek V4.1 Flash 1017 or GPT-5?
DeepSeek V4.1 Flash 1017, with 1.05M tokens against 400K.
How many benchmarks is the DeepSeek V4.1 Flash 1017 vs GPT-5 comparison based on?
21 benchmarks have a result for both models as of October 22, 2026. DeepSeek V4.1 Flash 1017's coding index rests on 4 independent benchmarks and GPT-5's on 3. Where only one model has a result, the page says so instead of naming a winner.