DeepSeek V4 Flash 0731 vs GPT-6 Luna
GPT-6 Luna leads DeepSeek V4 Flash 0731 on coding, 57.3 to 51.8 on the Vibe Coding Index, and DeepSeek V4 Flash 0731 costs 1.3x less per output token ($0.396 vs $0.50 per million). Updated October 22, 2026; 42 benchmarks have a result for both.
DeepSeek V4 Flash 0731 or GPT-6 Luna: which to pick
Pick GPT-6 Luna for coding: it scores 57.3 on the Vibe Coding Index against 51.8 for DeepSeek V4 Flash 0731.
Pick DeepSeek V4 Flash 0731 for output tokens 1.3x cheaper ($0.396 vs $0.50 per million) and work measured by Terminal-Bench 4 (Vals AI) or Tax Agent Bench (Vals AI), where it leads.
- Shared results
- 42
- DeepSeek V4 Flash 0731 only
- 13
- GPT-6 Luna only
- 61
- Like-for-like categories
- 8 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
GPT-6 Luna
GPT-6 Luna scores 57.3 against 51.8.
4 and 5 benchmarks behind the scores
Intelligence
AA Intelligence Index
GPT-6 Luna
GPT-6 Luna leads, 38.1 against 34.3.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only DeepSeek V4 Flash 0731 has a AA Agentic Index result, 41.0.
Directional only
Reasoning
GPQA Diamond
No call
Only DeepSeek V4 Flash 0731 has a GPQA Diamond result, 89.9%.
Directional only
Knowledge
MMLU Pro
No call
Only DeepSeek V4 Flash 0731 has a MMLU Pro result, 86.2%.
Directional only
Web design
Design Arena: websites
GPT-6 Luna
GPT-6 Luna leads, 1271 against 1237.
Same board, both measured
General
Vals Index
No call
Only GPT-6 Luna has a Vals Index result, 52.3%.
Directional only
Long documents
Listed context window
GPT-6 Luna
GPT-6 Luna takes 1.05M tokens against 1.05M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V4 Flash 0731
DeepSeek V4 Flash 0731 $0.0003, GPT-6 Luna $0.0003 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only DeepSeek V4 Flash 0731 has the number, $0.01.
Directional only
At a glance
- Coding index
- DeepSeek V4 Flash 0731 51.8 from 4 benchmarks, GPT-6 Luna 57.3 from 5 benchmarks.
- Widest category
- Math, 16.4 points apart on average over 4 shared benchmarks.
- Widest single gap
- FrontierCode: DeepSeek V4 Flash 0731 18.8% against GPT-6 Luna 42.4%, 36.3 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V4 Flash 0731 $0.396 against GPT-6 Luna $0.50 per million output tokens, 1.3x apart.
- Context window
- DeepSeek V4 Flash 0731 takes 1.05M tokens, GPT-6 Luna 1.05M.
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 Flash 0731 | Bars | GPT-6 Luna | Ties |
|---|---|---|---|---|
| Coding | 1 | 9 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 2 | 2 | 2 ties | |
| Reasoning | 5 | 1 | ||
| Math | 0 | 4 | ||
| Knowledge | 0 | 2 | ||
| Web design | 0 | 8 | ||
| General | 1 | 3 | ||
| All shared boards | 9 | 31 | 2 ties |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index19
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 84.5% | Only GPT-6 Luna measured |
| LMArena WebDev | Not measured | 1578 | Only GPT-6 Luna measured |
| AA Coding Index | 69.1 | Not measured | Only DeepSeek V4 Flash 0731 measured |
| LiveCodeBench (Vals AI) | 87.3% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| SWE-bench Verified (Vals AI) | 88.8% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| Terminal-Bench 4 (Vals AI) | 18.7% | 13.6% | DeepSeek V4 Flash 0731 |
| Vibe Code Bench (Vals AI) | 74.7% | 81.7% | GPT-6 Luna |
| Code migration (Vals AI) | 38.6% | 42.5% | GPT-6 Luna |
| ProgramBench (Vals AI) | 0.0% | 0.5% | GPT-6 Luna |
| IOI (Vals AI) | 32.7% | 55.6% | GPT-6 Luna |
| Terminal-Bench 2.1 (Vals AI) | 67.0% | 73.0% | GPT-6 Luna |
| SWE-bench Multimodal (Vals AI) | Not measured | 37.5% | Only GPT-6 Luna measured |
| SkillsBench (Vals AI) | 50.7% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| CyberBench (Vals AI) | 72.1% | 76.3% | GPT-6 Luna |
| SciCode | 50.3% | 54.6% | GPT-6 Luna |
| FrontierCode | 18.8% | 42.4% | GPT-6 Luna |
| WeirdML | 63.0% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| ALE-Bench | 1306.1 | 1576.9 | GPT-6 Luna |
| LMArena Text: coding | Not measured | 1437 | Only GPT-6 Luna measured |
| LMArena Text: software and IT | Not measured | 1423 | Only GPT-6 Luna measured |
| LMArena WebDev: React | Not measured | 1573 | Only GPT-6 Luna measured |
| LMArena WebDev: HTML | Not measured | 1595 | Only GPT-6 Luna measured |
| LMArena WebDev: from an image | Not measured | 1571 | Only GPT-6 Luna measured |
| Neither model has a result on the other 16 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| AA Intelligence Index | 34.3 | 38.1 | GPT-6 Luna |
| Epoch Capabilities Index | 154.8 | 156.8 | GPT-6 Luna |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index22
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| AA Agentic Index | 41.0 | Not measured | Only DeepSeek V4 Flash 0731 measured |
| Design Arena: agents building web apps | Not measured | 1191 | Only GPT-6 Luna measured |
| Design Arena: agents building full-stack apps | Not measured | 1158 | Only GPT-6 Luna measured |
| Design Arena: agents building mobile apps | Not measured | 1134 | Only GPT-6 Luna measured |
| LiveBench research | Not measured | 43.5% | Only GPT-6 Luna measured |
| LiveBench browser | Not measured | 72.0% | Only GPT-6 Luna measured |
| LiveBench long agentic | Not measured | 48.1% | Only GPT-6 Luna measured |
| SREBench (Vals AI) | Not measured | 2.7% | Only GPT-6 Luna measured |
| Finance Agent (Vals AI) | 49.5% | 49.9% | GPT-6 Luna |
| Tax Agent Bench (Vals AI) | 28.1% | 19.9% | DeepSeek V4 Flash 0731 |
| Legal Research Bench (Vals AI) | 30.3% | 30.3% | Tie |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 4.2% | Only GPT-6 Luna measured |
| Excel Modeling Benchmark (Vals AI) | 57.0% | 68.5% | GPT-6 Luna |
| BioMysteryBench (Vals AI) | 64.4% | 61.5% | DeepSeek V4 Flash 0731 |
| Terminal-Bench-Science (Vals AI) | 4.3% | 4.3% | Tie |
| Vals Science Index | Not measured | 28.4% | Only GPT-6 Luna measured |
| APEX-Agents | Not measured | 44.3% | Only GPT-6 Luna measured |
| SpireClimb | Not measured | 0.0% | Only GPT-6 Luna measured |
| Neither model has a result on the other 23 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index51
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| GPQA Diamond | 89.9% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| LiveBench reasoning | Not measured | 81.8% | Only GPT-6 Luna measured |
| MysteryMechanism (Vals AI) | Not measured | 19.4% | Only GPT-6 Luna measured |
| GPQA Diamond (Artificial Analysis run) | 90.8% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| Humanity's Last Exam (Artificial Analysis run) | 38.6% | 38.5% | DeepSeek V4 Flash 0731 |
| GPQA Diamond (Epoch AI run) | 91.0% | 90.5% | DeepSeek V4 Flash 0731 |
| ARC-AGI | 89.0% | 86.7% | DeepSeek V4 Flash 0731 |
| ARC-AGI-2 | 61.4% | 59.3% | DeepSeek V4 Flash 0731 |
| SimpleBench | 61.1% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| Chess puzzles | 32.0% | 36.0% | GPT-6 Luna |
| Mystery game puzzles | 34.0% | 7.0% | DeepSeek V4 Flash 0731 |
| LMArena Text: hard prompts | Not measured | 1410 | Only GPT-6 Luna measured |
| LMArena Text: expert | Not measured | 1442 | Only GPT-6 Luna measured |
| LMArena Text: hard prompts in English | Not measured | 1410 | Only GPT-6 Luna measured |
| LMArena Vision: diagrams | Not measured | 1245 | Only GPT-6 Luna measured |
| LMArena Vision: homework | Not measured | 1261 | Only GPT-6 Luna measured |
| Neither model has a result on the other 6 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| OTIS Mock AIME | 94.4% | 98.9% | GPT-6 Luna |
| FrontierMath, tiers 1 to 3 | 57.5% | 78.9% | GPT-6 Luna |
| FrontierMath, tier 4 | 24.4% | 56.1% | GPT-6 Luna |
| ProofBench | 56.0% | 64.0% | GPT-6 Luna |
| LMArena Text: math | Not measured | 1409 | Only GPT-6 Luna measured |
| LMArena Text: mathematical work | Not measured | 1427 | Only GPT-6 Luna measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| MMLU Pro | 86.2% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| LegalBench (Vals AI) | 77.7% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| CorpFin (Vals AI) | 61.9% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| TaxEval (Vals AI) | 70.7% | Not measured | Only DeepSeek V4 Flash 0731 measured |
| MedCode (Vals AI) | 41.4% | 44.7% | GPT-6 Luna |
| SimpleQA Verified | 33.6% | 41.4% | GPT-6 Luna |
| LMArena Text: science | Not measured | 1386 | Only GPT-6 Luna measured |
| LMArena Text: medicine and healthcare | Not measured | 1352 | Only GPT-6 Luna measured |
| LMArena Text: legal and government | Not measured | 1400 | Only GPT-6 Luna measured |
| LMArena Text: business and finance | Not measured | 1388 | Only GPT-6 Luna measured |
| FACTS Benchmark Suite | Not measured | 63.3% | Only GPT-6 Luna measured |
| Neither model has a result on the other 7 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| Design Arena: websites | 1237 | 1271 | GPT-6 Luna |
| Design Arena: UI components | 1233 | 1295 | GPT-6 Luna |
| Design Arena: game dev | 1207 | 1257 | GPT-6 Luna |
| Design Arena: data viz | 1180 | 1216 | GPT-6 Luna |
| Design Arena: 3D | 1208 | 1218 | GPT-6 Luna |
| Design Arena: all code categories | 1229 | 1270 | GPT-6 Luna |
| Design Arena: SVG | 1184 | 1252 | GPT-6 Luna |
| Design Arena: ASCII art | 1090 | 1153 | GPT-6 Luna |
| LiveBench design | Not measured | 82.5% | Only GPT-6 Luna measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index13
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | Leader |
|---|---|---|---|
| Vals Index | Not measured | 52.3% | Only GPT-6 Luna measured |
| LiveBench data analysis | Not measured | 73.4% | Only GPT-6 Luna measured |
| Public Benefits Bench (Vals AI) | Not measured | 57.6% | Only GPT-6 Luna measured |
| SAGE (Vals AI) | Not measured | 48.1% | Only GPT-6 Luna measured |
| MedScribe (Vals AI) | 80.4% | 83.7% | GPT-6 Luna |
| Long context reasoning (AA-LCR) | 79.7% | 83.3% | GPT-6 Luna |
| DTBench | 90.9% | 90.1% | DeepSeek V4 Flash 0731 |
| LMCA | 41.7% | 44.5% | GPT-6 Luna |
| Furniture assembly | Not measured | 44.2% | Only GPT-6 Luna measured |
| LMArena Text | Not measured | 1389 | Only GPT-6 Luna measured |
| LMArena Text: instruction following | Not measured | 1408 | Only GPT-6 Luna measured |
| LMArena Text: creative writing | Not measured | 1363 | Only GPT-6 Luna measured |
| LMArena Vision | Not measured | 1216 | Only GPT-6 Luna measured |
| LMArena Text: English | Not measured | 1393 | Only GPT-6 Luna measured |
| LMArena Text: non-English | Not measured | 1381 | Only GPT-6 Luna measured |
| LMArena Text: Chinese | Not measured | 1427 | Only GPT-6 Luna measured |
| LMArena Text: German | Not measured | 1376 | Only GPT-6 Luna measured |
| LMArena Text: Japanese | Not measured | 1369 | Only GPT-6 Luna measured |
| LMArena Text: Korean | Not measured | 1359 | Only GPT-6 Luna measured |
| LMArena Text: French | Not measured | 1408 | Only GPT-6 Luna measured |
| LMArena Text: Spanish | Not measured | 1391 | Only GPT-6 Luna measured |
| LMArena Text: Russian | Not measured | 1392 | Only GPT-6 Luna measured |
| LMArena Text: Polish | Not measured | 1383 | Only GPT-6 Luna measured |
| LMArena Text: multi-turn | Not measured | 1392 | Only GPT-6 Luna measured |
| LMArena Text: longer queries | Not measured | 1408 | Only GPT-6 Luna measured |
| LMArena Text: writing and language | Not measured | 1384 | Only GPT-6 Luna measured |
| LMArena Text: entertainment and media | Not measured | 1364 | Only GPT-6 Luna measured |
| LMArena Vision: English | Not measured | 1204 | Only GPT-6 Luna measured |
| LMArena Vision: Chinese | Not measured | 1266 | Only GPT-6 Luna measured |
| LMArena Vision: reading text | Not measured | 1225 | Only GPT-6 Luna measured |
| LMArena Vision: creative writing | Not measured | 1205 | Only GPT-6 Luna measured |
| Neither model has a result on the other 15 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V4 Flash 0731 | GPT-6 Luna | 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 Flash 0731 costs less
Repository review
50K input + 3K output tokens
GPT-6 Luna costs less
Agent loop
200K input + 10K output tokens
GPT-6 Luna 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 Flash 0731 | OpenAI GPT-6 Luna |
|---|---|---|
| Maker | DeepSeek | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | July 31, 2026 | September 22, 2026 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $0.14 | $0.10 |
| Output $/Mtok | $0.396 | $0.50 |
| Coding index | 51.8 | 57.3 |
| Independent measurements behind it | 4 | 5 |
| Cost per solved task | $0.01 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V4 Flash 0731 better than GPT-6 Luna for coding?
Pick GPT-6 Luna for coding: it scores 57.3 on the Vibe Coding Index against 51.8 for DeepSeek V4 Flash 0731. Pick DeepSeek V4 Flash 0731 for output tokens 1.3x cheaper ($0.396 vs $0.50 per million) and work measured by Terminal-Bench 4 (Vals AI) or Tax Agent Bench (Vals AI), where it leads.
Which is cheaper, DeepSeek V4 Flash 0731 or GPT-6 Luna?
DeepSeek V4 Flash 0731 costs $0.14 per million input tokens and $0.396 per million output tokens; GPT-6 Luna costs $0.10 and $0.50. Cost per solved task is known for DeepSeek V4 Flash 0731 only.
Which has the bigger context window, DeepSeek V4 Flash 0731 or GPT-6 Luna?
GPT-6 Luna, with 1.05M tokens against 1.05M.
How many benchmarks is the DeepSeek V4 Flash 0731 vs GPT-6 Luna comparison based on?
42 benchmarks have a result for both models as of October 22, 2026. DeepSeek V4 Flash 0731's coding index rests on 4 independent benchmarks and GPT-6 Luna's on 5. Where only one model has a result, the page says so instead of naming a winner.