Compare / DeepSeek-V3.1 vs DeepSeek-V3.2
DeepSeek-V3.1 vs DeepSeek-V3.2
Which is cheaper, which is more capable, and which does better at each kind of work, from 18 results on 4 sources that measured both models.
Of the 18 results both models have, the better value on each. Results whose reported ranges overlap are too close to call.
Results as of 16 November 2026 (catalogue release 2026-11-16-49bdfa045dee).
DeepSeek-V3.1 or DeepSeek-V3.2? The short answer
Pick DeepSeek-V3.2: it scores higher (Intelligence Index 21.5 against 13.7) and costs less.
- DeepSeek-V3.2 is cheaper: $0.32 against $0.45 per million tokens, 30% less.
- DeepSeek-V3.2 scores higher on the Artificial Analysis Intelligence Index: 21.5 against 13.7.
- DeepSeek-V3.2 is newer: released 1 Dec 2025, against 21 Aug 2025 for DeepSeek-V3.1.
- On third-party benchmarks, on the results both models have: DeepSeek-V3.1 does better for sticking to the facts; DeepSeek-V3.2 for reasoning and knowledge and writing.
Head to head
Every fact side by side
| DeepSeek-V3.1 | DeepSeek-V3.2 | |
|---|---|---|
| Price per million tokens | $0.27 in · $1.00 out | $0.28 in · $0.42 out |
| Blended (3 in : 1 out) | $0.45 | $0.32 |
| Cheapest host, blended | $0.42 (DeepInfra, FP4) | $0.23 (GMICloud, FP8) |
| Intelligence Index | 13.7 | 21.5 |
| Spring Prompt overall | – | – |
| Output speed, tokens a second | – | – |
| Context window | 163,840 tokens | 163,840 tokens |
| Released | 21 Aug 2025 | 1 Dec 2025 (newer) |
| Weights | Open | 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 | DeepSeek-V3.1 | DeepSeek-V3.2 |
|---|---|---|
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results | ||
| Overall · TextArena rating, higher is better 116 · 106 of 387 measured | 1,417 | 1,425 |
| Business, management and finance · TextArena rating, higher is better 112 · 102 of 380 measured | 1,417thinking reasoning | 1,423thinking reasoning |
| Creative writing · TextArena rating, higher is better 91 · 93 of 385 measured | 1,402thinking reasoning | 1,401 |
| Artificial Analysis reported by Artificial Analysis · 3 results | ||
| Artificial Analysis Intelligence Indexindex score, higher is better 142 · 102 of 249 measured | 13.7no reasoning | 21.5thinking reasoning |
| GPQA Diamond% of questions, higher is better 122 · 90 of 232 measured | 77.9%thinking reasoning | 84.0%thinking reasoning |
| Humanity's Last Exam% of questions, higher is better 138 · 101 of 250 measured | 14.3%thinking reasoning | 24.6%thinking reasoning |
| UGI Leaderboard reported by UGI Leaderboard · 3 results | ||
| Writing score · UGIscore out of 100, higher is better 73 · 54 of 230 measured | 51.9thinking reasoning | 57.3thinking reasoning |
| Requested-length error · UGI% off the requested word count, lower is better 166 · 151 of 230 measured | 30.0%no reasoning | 26.0%no reasoning |
| Style adherence · UGIscore from 0 to 1, higher is better 66 · 55 of 230 measured | 0.36thinking reasoning | 0.37thinking reasoning |
| Vectara Hallucination Leaderboard reported by Vectara Hallucination Leaderboard · 2 results | ||
| Hallucination rate · Vectara% of summaries, lower is better 18 · 27 of 104 measured | 5.5% | 6.3% |
| Answer rate · Vectara% of documents, higher is better 89 · 94 of 104 measured | 94.5% | 92.6% |
Show all 18 results (7 more)
| Measure | DeepSeek-V3.1 | DeepSeek-V3.2 |
|---|---|---|
| Arena (formerly LMArena) reported by Arena (formerly LMArena) · 3 results | ||
| Expert prompts · TextArena rating, higher is better 122 · 94 of 338 measured | 1,432thinking reasoning | 1,452thinking reasoning |
| Instruction following · TextArena rating, higher is better 97 · 94 of 387 measured | 1,416thinking reasoning | 1,420 |
| Writing, literature and language · TextArena rating, higher is better 94 · 91 of 386 measured | 1,407thinking reasoning | 1,410 |
| Artificial Analysis reported by Artificial Analysis · 4 results | ||
| IFBench% of instructions, higher is better 127 · 79 of 195 measured | 41.5%thinking reasoning | 60.7%thinking reasoning |
| Long-context reasoning (AA-LCR)% of questions, higher is better 149 · 98 of 241 measured | 56.7%thinking reasoning | 73.3%thinking reasoning |
| Terminal-Bench Hard% of tasks, higher is better 78 · 47 of 190 measured | 25.0%thinking reasoning | 35.6%thinking reasoning |
| τ²-bench telecom% of tasks, higher is better 117 · 40 of 193 measured | 37.4%thinking reasoning | 90.6%thinking reasoning |
Quick answers
Is DeepSeek-V3.1 better than DeepSeek-V3.2?
Pick DeepSeek-V3.2: it scores higher (Intelligence Index 21.5 against 13.7) and costs less.
Which is cheaper, DeepSeek-V3.1 or DeepSeek-V3.2?
DeepSeek-V3.2 is cheaper: $0.32 against $0.45 per million tokens, 30% less.
Which is better for reasoning and knowledge, DeepSeek-V3.1 or DeepSeek-V3.2?
DeepSeek-V3.2 ranks 93 of 187 models on sale for hard questions across science, maths and general knowledge, against 120 for DeepSeek-V3.1.
Which is better for writing, DeepSeek-V3.1 or DeepSeek-V3.2?
DeepSeek-V3.2 ranks 65 of 169 models on sale for what people prefer in blind comparisons, and judged writing quality, against 83 for DeepSeek-V3.1.
Which is better for sticking to the facts, DeepSeek-V3.1 or DeepSeek-V3.2?
DeepSeek-V3.1 ranks 19 of 139 models on sale for summarising without inventing things, and factual answers, against 58 for DeepSeek-V3.2.
Run DeepSeek-V3.1 and DeepSeek-V3.2 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.