Compare / DeepSeek-V4.1-Flash-1017 vs GPT-6 Luna

DeepSeek-V4.1-Flash-1017 vs GPT-6 Luna

Which is cheaper, which is more capable, and which does better at each kind of work, from 90 results on 6 sources that measured both models.

DeepSeekDeepSeek-V4.1-Flash-1017 Released 17 Oct 2026 · $0.30 in · $1.20 out
OpenAIGPT-6 Luna Released 22 Sep 2026 · $0.10 in · $0.50 out
DeepSeek-V4.1-Flash-1017 wins 1721 too close to callGPT-6 Luna wins 52

Of the 90 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.

Results as of 19 November 2026 (catalogue release 2026-11-19-3608133e8331). See also DeepSeek vs OpenAI, every model from both labs.

DeepSeek-V4.1-Flash-1017 or GPT-6 Luna? The short answer

They score about the same, so pick GPT-6 Luna: it costs 3.2 times less per product listing on CatalogBench. Pick DeepSeek-V4.1-Flash-1017 if speed matters: it writes faster.

  • Spring Prompt overall: DeepSeek-V4.1-Flash-1017 37, GPT-6 Luna 38 (out of 100; how the score works).
  • GPT-6 Luna is cheaper to run: $0.000 against $0.002 per product listing on CatalogBench (3.2 times less).
  • DeepSeek-V4.1-Flash-1017 scores higher on the Artificial Analysis Intelligence Index: 45.1 against 38.1.
  • DeepSeek-V4.1-Flash-1017 writes faster: 212 tokens a second against 127.
  • DeepSeek-V4.1-Flash-1017 is newer: released 17 Oct 2026, against 22 Sep 2026 for GPT-6 Luna.
  • On our own benchmarks, on the results both models have: GPT-6 Luna does better for product listings, decks from an analysis and marketing planning.
  • On third-party benchmarks, on the results both models have: DeepSeek-V4.1-Flash-1017 does better for agents and tool use and reasoning and knowledge.

Head to head

37
Spring Prompt overall
38
45.1
Intelligence Index
38.1
$0.52
Price per million tokens, blendedlower is better
$0.20
212
Output speed, tokens a second
127
1,048,576
Context window, tokens
1,050,000

Every fact side by side

DeepSeek-V4.1-Flash-1017GPT-6 Luna
Price per million tokens$0.30 in · $1.20 out$0.10 in · $0.50 out
Blended (3 in : 1 out)$0.52$0.20
Cheapest host, blended$0.11 (Sail Research, FP4)$0.20 (Amazon Bedrock)
Cost per product listing (CatalogBench)$0.002$0.000
Cached input, per million$0.006 cached input$0.010 cached input
Intelligence Index45.138.1
Spring Prompt overall3738
Output speed, tokens a second212127
Context window1,048,576 tokens1,050,000 tokens
Released17 Oct 2026 (newer)22 Sep 2026
WeightsOpenClosed
Developer based inChinathe US

Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.

Which is better for what

Each model's rank among the models on sale today, for each kind of work, from our own benchmarks and licensed sources. Further right is better; the better rank is in green.

Product listingsTurning a sparse product feed and photos into listings that can go live 31 of 35 10 of 35
Decks from an analysisTurning a finished analysis into a deck you could present as it is 22 of 35 12 of 35
User surveysPlanning a user survey and reading its results without being misled 12 of 26 14 of 26
Marketing planningPlanning a year of ad spend without overspending 23 of 35 7 of 35
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 24 of 159 53 of 159
Professional workReal tasks from banking, consulting and law, business documents and freelance projects not measured 25 of 51
Reasoning and knowledgeHard questions across science, maths and general knowledge 24 of 187 48 of 187
WritingWhat people prefer in blind comparisons, and judged writing quality not measured 61 of 169
Sticking to the factsSummarising without inventing things, and factual answers not measured 89 of 139
Speed under pressureGood decisions against a real clock (fast chess) 29 of 38 28 of 38

From our research

Every shared result

Each model's best result on every measure both have, grouped by benchmark, with the setting that produced it. Under each measure, every model that benchmark has measured is a grey tick, best to the right, so you can see where the two sit in the field. The headline measures of each benchmark come first; the rest are under "Show all". Green marks the better value; where a source reports ranges that overlap, neither is marked.

MeasureDeepSeek-V4.1-Flash-1017GPT-6 Luna
BulletBench measured by Spring Prompt · 3 results
Cost per game · Bullet 60sUS dollars, lower is better 32 · 21 of 62 measured $0.0047low reasoning $0.0016no reasoning
Ladder Elo · Bullet 60sladder Elo, higher is better 54 · 52 of 62 measured 147low reasoning 249no reasoning
Games lost on time · Bullet 60s% of games, lower is better 57 · 39 of 62 measured 87.5%low reasoning 50.0%no reasoning
CatalogBench measured by Spring Prompt · 3 results
Cost per productUS dollars, lower is better 5 · 2 of 36 measured $0.0015 $0.0005
Reliably publish-ready% of products, higher is better 28 · 13 of 35 measured 21.4% 55.4%
Reliably publish-ready · sales brief% of products, higher is better 30 · 7 of 35 measured 0.0% 51.8%
DeckBench measured by Spring Prompt · 3 results
Cost per deckUS dollars, lower is better 9 · 3 of 36 measured $0.0255 $0.0087
Deck ratingrating, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. 22 · 12 of 36 measured 1,046 1,252
Accurate decks% of tasks, higher is better 22 · 17 of 36 measured 33.3% 50.0%
ROASBench measured by Spring Prompt · 3 results
Overall scorescore out of 100, higher is better 23 · 7 of 36 measured 41.4 53.4
Audience scorescore out of 100, higher is better 17 · 7 of 36 measured 65.3 68.3
Business scorescore out of 100, higher is better 23 · 7 of 36 measured 38.3 54.8
SurveyBench measured by Spring Prompt · 3 results
Analysis ratingrating, higher is betterThe rating weighs each win by how strong the other analysis was; the win rate counts wins against whichever analyses a model happened to meet, so the two can disagree. 12 · 14 of 27 measured 1,119 1,072
Cost per taskUS dollars, lower is better 8 · 1 of 27 measured $0.0373 $0.0088
Analysis scorepoints of 100, higher is better 8 · 7 of 27 measured 90.2 90.8
Artificial Analysis reported by Artificial Analysis · 3 results
Artificial Analysis Intelligence Indexindex score, higher is better 16 · 39 of 249 measured 45.1max reasoning 38.1max reasoning
Humanity's Last Exam% of questions, higher is better 29 · 53 of 250 measured 43.3%max reasoning 38.5%max reasoning
Terminal-Bench 4.0% of tasks, higher is better 18 · 37 of 120 measured 33.8%max reasoning 12.6%max reasoning
Show all 90 results (72 more)
MeasureDeepSeek-V4.1-Flash-1017GPT-6 Luna
BulletBench measured by Spring Prompt · 2 results
Invalid moves · Bullet 60s% of moves, lower is better 1 · 1 of 62 measured 0.0%low reasoning 0.0%no reasoning
Median move time · Bullet 60smilliseconds, lower is better 59 · 39 of 62 measured 3.3 slow reasoning 1.5 sno reasoning
CatalogBench measured by Spring Prompt · 28 results
Channel compliance% of products, higher is better 1 · 1 of 36 measured 100.0% 100.0%
Channel rules broken% of products, lower is better 30 · 1 of 36 measured 1.2% 0.0%
Channel rules broken · sales brief% of products, lower is better 33 · 1 of 36 measured 2.4% 0.0%
Claims to check% of products, lower is better 28 · 1 of 34 measured 1.2% 0.0%
Claims to check · sales brief% of products, lower is better 28 · 1 of 34 measured 3.6% 0.0%
Conflicts caught% of conflicts, higher is better 1 · 1 of 36 measured 100.0% 100.0%
Content quality% of checks, higher is better 32 · 20 of 36 measured 95.3% 97.0%
Content quality · sales brief% of checks, higher is better 31 · 18 of 36 measured 94.8% 97.4%
Decision accuracy% of decisions, higher is better 30 · 16 of 36 measured 94.9% 97.0%
Failed outputs% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Failed outputs · sales brief% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Field accuracy% of missing fields, higher is better 18 · 18 of 36 measured 93.7% 93.7%
Invented values% of filled values, lower is better 18 · 18 of 36 measured 1.7% 1.7%
Missing UK information% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Missing UK information · sales brief% of products, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Not findable% of products, lower is better 13 · 32 of 36 measured 3.6% 10.1%
Not findable · sales brief% of products, lower is better 12 · 32 of 36 measured 2.4% 10.1%
Publish-ready listings% of products, higher is better 29 · 17 of 36 measured 40.5% 67.9%
Publish-ready listings · sales brief% of products, higher is better 31 · 8 of 36 measured 8.3% 67.9%
Time per productseconds, lower is better 4 · 8 of 36 measured 6.70 8.20
Unsupported claimsclaims per product, lower is better 33 · 13 of 36 measured 1.70 0.31
Unsupported claims% of products, lower is better 32 · 13 of 36 measured 44.6% 8.9%
Unsupported claims · sales briefclaims per product, lower is better 34 · 8 of 36 measured 4.01 0.35
Unsupported claims · sales brief% of products, lower is better 35 · 8 of 36 measured 85.7% 8.3%
Wrong attributes% of products, lower is better 23 · 24 of 36 measured 17.9% 18.5%
Wrong attributes · sales brief% of products, lower is better 32 · 23 of 36 measured 22.0% 19.1%
Wrong category or variant% of products, lower is better 2 · 7 of 36 measured 0.6% 1.8%
Wrong category or variant · sales brief% of products, lower is better 5 · 1 of 36 measured 1.2% 0.0%
DeckBench measured by Spring Prompt · 14 results
Caveat dropped% of tasks, lower is better 16 · 1 of 36 measured 16.7% 0.0%
Clean layout% of tasks, higher is better 26 · 10 of 36 measured 16.7% 66.7%
Design quality% of the maximum, higher is better 15 · 10 of 36 measured 68.5% 78.1%
Draft figure quoted% of tasks, lower is better 29 · 21 of 36 measured 50.0% 33.3%
Findings missing% of tasks, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other deck was; the win rate counts wins against whichever decks a model happened to meet, so the two can disagree. 22 · 9 of 36 measured 46.0% 70.9%
Layout defects% of tasks, lower is better 26 · 10 of 36 measured 83.3% 33.3%
Misleading metric used% of tasks, lower is better 1 · 1 of 36 measured 0.0% 0.0%
Presentable decks% of tasks, higher is better 17 · 17 of 36 measured 0.0% 0.0%
Recommendation late or wrong% of tasks, lower is better 17 · 17 of 36 measured 16.7% 16.7%
Slides needing work% of tasks, lower is better 26 · 5 of 36 measured 66.7% 16.7%
Time per deckseconds, lower is better 9 · 17 of 36 measured 73.0 101
Unsupported claims% of tasks, lower is better 17 · 1 of 36 measured 16.7% 0.0%
Unsupported numbers% of tasks, lower is better 1 · 1 of 36 measured 0.0% 0.0%
ROASBench measured by Spring Prompt · 6 results
Consistency scorescore out of 100, higher is better 27 · 10 of 36 measured 20.2 35.8
Contribution profitsimulated US dollars, higher is better 20 · 9 of 36 measured $585,572 $738,879
Cost of a runUS dollars, lower is better 10 · 2 of 36 measured $0.0799 $0.0142
Months over budgetmonths of 12, lower is better 1 · 1 of 36 measured 0 0
Planning scorescore out of 100, higher is better 29 · 20 of 36 measured 54.3 54.8
Return on ad spendprofit per $1 spent, higher is better 24 · 10 of 36 measured $0.44 $0.66
SurveyBench measured by Spring Prompt · 18 results
Head-to-head win rate% of comparisons, higher is betterThe rating weighs each win by how strong the other analysis was; the win rate counts wins against whichever analyses a model happened to meet, so the two can disagree. 12 · 17 of 27 measured 55.2% 47.1%
Leading or double-barrelled questions% of tasks, lower is better 5 · 5 of 27 measured 11.1% 11.1%
Mistaken next steps% of tasks, lower is better 1 · 5 of 27 measured 0.0% 11.1%
Numbers right% of answers, higher is better 1 · 21 of 27 measured 100.0% 98.0%
Objectives unanswerable% of tasks, lower is better 18 · 1 of 27 measured 11.1% 0.0%
Output tokens per tasktokens, lower is better 8 · 4 of 17 measured 30,893 14,333
Precision ignored% of tasks, lower is better 1 · 1 of 27 measured 0.0% 0.0%
Research scorepoints of 100, higher is better 8 · 7 of 27 measured 90.9 91.7
Sound research% of tasks, higher is better 6 · 6 of 27 measured 33.3% 33.3%
Stakeholder's view presumed% of tasks, lower is better 1 · 12 of 27 measured 0.0% 11.1%
Survey plan scorepoints of 100, higher is better 14 · 9 of 27 measured 92.5 94.0
Survey structure% of tasks, lower is better 22 · 12 of 27 measured 33.3% 22.2%
Time per taskseconds, lower is better 11 · 12 of 27 measured 134 138
Traps handled% of traps, higher is better 16 · 11 of 27 measured 91.3% 94.2%
Traps missed% of tasks, lower is better 13 · 8 of 27 measured 44.4% 33.3%
Unsupported findings% of tasks, lower is better 11 · 7 of 27 measured 44.4% 22.2%
Wrong numbers% of tasks, lower is better 1 · 14 of 27 measured 0.0% 11.1%
Wrong recommendation% of tasks, lower is better 5 · 5 of 27 measured 11.1% 11.1%
Artificial Analysis reported by Artificial Analysis · 4 results
Long-context reasoning (AA-LCR)% of questions, higher is better 21 · 13 of 241 measured 82.7%max reasoning 83.3%max reasoning
Output speedtokens per second, higher is better 15 · 37 of 78 measured 212max reasoning 127max reasoning
SciCode% of problems, higher is better 25 · 31 of 121 measured 55.9%max reasoning 54.6%max reasoning
Time to first answer tokenseconds, lower is better 46 · 20 of 78 measured 10.5max reasoning 0.74no reasoning

Quick answers

Is DeepSeek-V4.1-Flash-1017 better than GPT-6 Luna?

They score about the same, so pick GPT-6 Luna: it costs 3.2 times less per product listing on CatalogBench. Pick DeepSeek-V4.1-Flash-1017 if speed matters: it writes faster.

Which is cheaper, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

GPT-6 Luna is cheaper to run: $0.000 against $0.002 per product listing on CatalogBench (3.2 times less).

Which is faster, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

DeepSeek-V4.1-Flash-1017 writes about 212 tokens a second on its usual API, against 127 for GPT-6 Luna (Artificial Analysis).

Which is better for product listings, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

GPT-6 Luna ranks 10 of 35 models on sale for turning a sparse product feed and photos into listings that can go live, against 31 for DeepSeek-V4.1-Flash-1017.

Which is better for decks from an analysis, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

GPT-6 Luna ranks 12 of 35 models on sale for turning a finished analysis into a deck you could present as it is, against 22 for DeepSeek-V4.1-Flash-1017.

Which is better for user surveys, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

DeepSeek-V4.1-Flash-1017 ranks 12 of 26 models on sale for planning a user survey and reading its results without being misled, against 14 for GPT-6 Luna.

Which is better for marketing planning, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

GPT-6 Luna ranks 7 of 35 models on sale for planning a year of ad spend without overspending, against 23 for DeepSeek-V4.1-Flash-1017.

Which has the bigger context window, DeepSeek-V4.1-Flash-1017 or GPT-6 Luna?

GPT-6 Luna: 1,050,000 tokens, against 1,048,576 for DeepSeek-V4.1-Flash-1017.

Run DeepSeek-V4.1-Flash-1017 and GPT-6 Luna on your own prompt.

Benchmarks aren't your task. Test both side by side on your prompt, with every answer checked against your rules and the cost of each one.