Compare / Qwen3.8-Max (0902) vs DeepSeek-V4-Pro (0423)
Qwen3.8-Max (0902) vs DeepSeek-V4-Pro (0423)
Which is cheaper, which is more capable, and which does better at each kind of work, from 42 results on 5 sources that measured both models.
Of the 42 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 13 October 2026 (catalogue release 2026-10-13-2e2420e3c5cc). See also Alibaba vs DeepSeek, every model from both labs.
Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)? The short answer
Pick Qwen3.8-Max (0902) if quality matters most: it scores higher (Spring Prompt overall 47 against 34). Pick DeepSeek-V4-Pro (0423) if cost matters more: it costs 3.1 times less per deck on DeckBench.
- Spring Prompt overall: Qwen3.8-Max (0902) 47, DeepSeek-V4-Pro (0423) 34 (out of 100; how the score works).
- DeepSeek-V4-Pro (0423) is cheaper to run: $0.046 against $0.15 per deck on DeckBench (3.1 times less).
- Qwen3.8-Max (0902) scores higher on the Artificial Analysis Intelligence Index: 45.4 against 30.4.
- Qwen3.8-Max (0902) is newer: released 3 Aug 2026, against 24 Apr 2026 for DeepSeek-V4-Pro (0423).
- On our own benchmarks, on the results both models have: Qwen3.8-Max (0902) does better for decks from an analysis; DeepSeek-V4-Pro (0423) for marketing planning.
- On third-party benchmarks, on the results both models have: Qwen3.8-Max (0902) does better for agents and tool use, reasoning and knowledge, writing and sticking to the facts.
Head to head
Every fact side by side
| Qwen3.8-Max (0902) | DeepSeek-V4-Pro (0423) | |
|---|---|---|
| Price per million tokens | $2.00 in · $6.00 out | $1.42 in · $2.83 out |
| Blended (3 in : 1 out) | $3.00 | $1.77 |
| Cheapest host, blended | – | $1.19 (StreamLake, FP8) |
| Cost per deck (DeckBench) | $0.15 | $0.046 |
| Intelligence Index | 45.4 | 30.4 |
| Spring Prompt overall | 47 | 34 |
| Output speed, tokens a second | 37 | – |
| Context window | 1,000,000 tokens | 1,048,576 tokens |
| Released | 3 Aug 2026 (newer) | 24 Apr 2026 |
| Weights | Closed | Open |
| Developer based in | China | China |
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.
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.
| Measure | Qwen3.8-Max (0902) | DeepSeek-V4-Pro (0423) |
|---|---|---|
| DeckBench measured by Spring Prompt · 3 results | ||
| Cost per deckUS dollars, lower is better 11 · 8 of 20 measured | $0.15 | $0.0465 |
| 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. 12 · 17 of 20 measured | 1,076 | 690 |
| Accurate decks% of tasks, higher is better 9 · 17 of 20 measured | 50.0% | 0.0% |
| ROASBench measured by Spring Prompt · 3 results | ||
| Overall scorescore out of 100, higher is better 15 · 14 of 20 measured | 25.4 | 28.3 |
| Audience scorescore out of 100, higher is better 18 · 17 of 20 measured | 39.7 | 44.8 |
| Business scorescore out of 100, higher is better 15 · 14 of 20 measured | 15.6 | 23.8 |
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results | ||
| Overall · TextArena rating, higher is better 20 · 53 of 387 measured | 1,482 | 1,458 |
| Overall · Text factualityArena rating, higher is better 22 · 49 of 163 measured | 1,472 | 1,455 |
| Business, management and finance · TextArena rating, higher is better 16 · 52 of 380 measured | 1,484 | 1,460 |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 14 · 49 of 243 measured | 45.4 | 30.4max reasoning |
| GPQA Diamond% of questions, higher is better 15 · 37 of 231 measured | 92.8% | 90.5%high reasoning |
| Humanity's Last Exam% of questions, higher is better 24 · 51 of 242 measured | 43.1% | 37.5%max reasoning |
| SimpleQA Verified (Epoch AI) reported by SimpleQA Verified (Epoch AI) · 1 result | ||
| Correct answers% of questions, higher is better 30 · 32 of 76 measured | 47.3%xhigh reasoning | 47.0%max reasoning |
Show all 42 results (29 more)
| Measure | Qwen3.8-Max (0902) | DeepSeek-V4-Pro (0423) |
|---|---|---|
| DeckBench measured by Spring Prompt · 13 results | ||
| Caveat dropped% of tasks, lower is better 11 · 11 of 20 measured | 16.7% | 16.7% |
| Clean layout% of tasks, higher is better 6 · 13 of 20 measured | 66.7% | 33.3% |
| Design quality% of the maximum, higher is better 10 · 19 of 20 measured | 67.0% | 40.9% |
| Draft figure quoted% of tasks, lower is better 12 · 12 of 20 measured | 33.3% | 33.3% |
| Findings missing% of tasks, lower is better 1 · 15 of 20 measured | 0.0% | 16.7% |
| 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. 11 · 18 of 20 measured | 61.3% | 30.4% |
| Layout defects% of tasks, lower is better 6 · 13 of 20 measured | 33.3% | 66.7% |
| Misleading metric used% of tasks, lower is better 1 · 1 of 20 measured | 0.0% | 0.0% |
| Presentable decks% of tasks, higher is better 5 · 12 of 20 measured | 33.3% | 0.0% |
| Recommendation late or wrong% of tasks, lower is better 1 · 17 of 20 measured | 0.0% | 66.7% |
| Slides needing work% of tasks, lower is better 10 · 18 of 20 measured | 50.0% | 83.3% |
| Unsupported claims% of tasks, lower is better 11 · 20 of 20 measured | 16.7% | 66.7% |
| Unsupported numbers% of tasks, lower is better 1 · 1 of 20 measured | 0.0% | 0.0% |
| ROASBench measured by Spring Prompt · 6 results | ||
| Consistency scorescore out of 100, higher is better 14 · 18 of 20 measured | 19.3 | 9.1 |
| Contribution profitsimulated US dollars, higher is better 15 · 13 of 20 measured | $99,775 | $301,289 |
| Cost of a runUS dollars, lower is better 16 · 10 of 20 measured | $0.73 | $0.20 |
| Months over budgetmonths of 12, lower is better 18 · 1 of 20 measured | 2 | 0 |
| Planning scorescore out of 100, higher is better 10 · 19 of 20 measured | 54.8 | 54.0 |
| Return on ad spendprofit per $1 spent, higher is better 15 · 13 of 20 measured | $0.09 | $0.26 |
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 4 results | ||
| Creative writing · TextArena rating, higher is better 17 · 39 of 385 measured | 1,468 | 1,446 |
| Expert prompts · TextArena rating, higher is better 20 · 64 of 338 measured | 1,513 | 1,480 |
| Instruction following · TextArena rating, higher is better 25 · 52 of 387 measured | 1,474 | 1,452 |
| Writing, literature and language · TextArena rating, higher is better 18 · 44 of 386 measured | 1,472 | 1,451 |
| Artificial Analysis reported by Artificial Analysis · 6 results | ||
| Terminal-Bench 2.1% of tasks, higher is better 5 · 41 of 130 measured | 88.8% | 64.8%high reasoning |
| Terminal-Bench 4.0% of tasks, higher is better 12 · 27 of 112 measured | 38.9% | 14.6%max reasoning |
| τ-bench banking% of tasks, higher is better 7 · 35 of 121 measured | 47.8% | 30.1%max reasoning |
| Artificial Analysis Coding Indexindex score, higher is better 10 · 40 of 131 measured | 76.2 | 59.4max reasoning |
| Long-context reasoning (AA-LCR)% of questions, higher is better 35 · 79 of 233 measured | 80.3% | 74.7%max reasoning |
| SciCode% of problems, higher is better 36 · 44 of 113 measured | 52.1% | 50.8%max reasoning |
Compare them with other models
- Alibaba vs DeepSeekEvery model from both labs, best against best
- Qwen3.8-Max (0902) vs Claude Opus 5.5Most intelligent at $3–10 per million tokens
- Qwen3.8-Max (0902) vs Claude Haiku 5.5Cheapest with similar intelligence
- DeepSeek-V4-Pro (0423) vs Muse Spark 1.3Most intelligent at $1–3 per million tokens
- DeepSeek-V4-Pro (0423) vs Claude Opus 5.5Most intelligent available today
Quick answers
Is Qwen3.8-Max (0902) better than DeepSeek-V4-Pro (0423)?
Pick Qwen3.8-Max (0902) if quality matters most: it scores higher (Spring Prompt overall 47 against 34). Pick DeepSeek-V4-Pro (0423) if cost matters more: it costs 3.1 times less per deck on DeckBench.
Which is cheaper, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
DeepSeek-V4-Pro (0423) is cheaper to run: $0.046 against $0.15 per deck on DeckBench (3.1 times less).
Which is better for decks from an analysis, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
Qwen3.8-Max (0902) ranks 12 of 20 models on sale for turning a finished analysis into a deck you could present as it is, against 17 for DeepSeek-V4-Pro (0423).
Which is better for marketing planning, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
DeepSeek-V4-Pro (0423) ranks 14 of 20 models on sale for planning a year of ad spend without overspending, against 15 for Qwen3.8-Max (0902).
Which is better for agents and tool use, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
Qwen3.8-Max (0902) ranks 7 of 164 models on sale for multi-step tasks with tools: support desks, coding agents, function calls, against 86 for DeepSeek-V4-Pro (0423).
Which is better for reasoning and knowledge, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
Qwen3.8-Max (0902) ranks 17 of 198 models on sale for hard questions across science, maths and general knowledge, against 45 for DeepSeek-V4-Pro (0423).
Which is better for writing, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
Qwen3.8-Max (0902) ranks 16 of 188 models on sale for what people prefer in blind comparisons, and judged writing quality, against 30 for DeepSeek-V4-Pro (0423).
Which has the bigger context window, Qwen3.8-Max (0902) or DeepSeek-V4-Pro (0423)?
DeepSeek-V4-Pro (0423): 1,048,576 tokens, against 1,000,000 for Qwen3.8-Max (0902).
Run Qwen3.8-Max (0902) and DeepSeek-V4-Pro (0423) on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.