Head to head
DeepSeek-V4-Pro vs DeepSeek-V3.2
DeepSeek-V4-Pro from DeepSeek against DeepSeek-V3.2 from DeepSeek — specification, price and every benchmark both makers have published, in one table.
At a glance
Benchmarks
No shared benchmarks
These two models have no benchmark in common with published figures for both, so there is nothing to compare directly. The specification and price rows below are still like for like.
Price
Not directly comparable
At least one of these does not have a published per-token price, so cost cannot be compared like for like. The open-weight model can be run on your own hardware instead.
What actually differs
- DeepSeek-V4-Pro takes 1M tokens of context against 164K — 6.4× more room for long documents or a large codebase.
Scorecard
Which is better at what
Maths, coding, reasoning and the rest — one line each, averaged over the benchmarks both models actually report.
DeepSeek-V4-Pro has no published benchmark scores
Its maker has not released figures for any of the evaluations tracked here, so there is nothing to put in a score column. Rather than estimate, infer from a sibling model, or quote the base model's numbers as if they were its own, this page leaves those rows empty and compares what genuinely can be compared: parameters, context window, modalities, licence and cost.
The moment those figures are published they go in — send them with a link to the source.
| Category | DeepSeek-V4-Pro | DeepSeek-V3.2 | Better at this |
|---|---|---|---|
| ReasoningOnly one model reports this | — | — | Not comparable |
| MathsOnly one model reports this | — | — | Not comparable |
| CodingOnly one model reports this | — | — | Not comparable |
| KnowledgeOnly one model reports this | — | — | Not comparable |
| MultimodalNeither model reports this | — | — | Not comparable |
| Instruction followingNeither model reports this | — | — | Not comparable |
| Human preferenceNeither model reports this | — | — | Not comparable |
| Categories won | No category has a test both models report | ||
Each category averages only the benchmarks every model here reports, so no one is credited for a test the other did not run. A category with no shared test is marked Not comparable rather than guessed at.
Side by side
DeepSeek-V4-Pro and DeepSeek-V3.2, row by row
| Attribute | DeepSeek-V4-ProDeepSeek | DeepSeek-V3.2DeepSeek |
|---|---|---|
| Specification | ||
| MakerWho built it | DeepSeek | DeepSeek |
| Released | 2026-04 | 2025-09 |
| ParametersTotal, and active per token for a mixture of experts | 1.6T total / 49B active | 685B total / 37B active |
| Architecture | MoE | MoE |
| Context windowHow much can go in at once | 1,048,576 tokens | 163,840 tokens |
| Max output | Not reported | 65,536 tokens |
| Input | Text | Text |
| ReasoningSpends extra tokens thinking before it answers | Not reported | Yes |
| Tool calling | Yes | Yes |
| Licence | MIT | MIT |
| Open weightsCan you download and run it yourself | Yes | Yes |
| Price | ||
| Input priceUSD per million tokens in | Not reported | $0.280 |
| Output priceUSD per million tokens out | Not reported | $0.42 |
| Blended 3:1A 3-in-to-1-out million-token mix — a fairer single number than input price alone | Not reported | $0.315 |
| Price note | First-party API rate. | Open weights — this is a representative hosting rate, not a first-party price. Running it yourself costs only hardware. |
| Published benchmarks | ||
| MMLU-Pro12,000 reasoning-heavy multiple-choice questions across 14 academic subjects, with ten options instead of four. The harder successor to MMLU. | Not reported | |
| GPQA Diamond198 graduate-level physics, chemistry and biology questions written to be Google-proof. PhD holders in the matching field score about 65%. | Not reported | |
| AIME 2025The American Invitational Mathematics Examination — 15 problems, integer answers, no partial credit. A standard test of multi-step maths reasoning. | Not reported | |
| LiveCodeBenchCompetitive-programming problems collected after each model's training cutoff, so contamination cannot inflate the score. | Not reported | |
| Links | Full page | Hugging Face · Full page |
| Row verified | 2026-10 | 2026-08 |
Where these numbers come from
Every score on this page is a published figure, taken from the model's own card, system card, technical report or release post, or from a public leaderboard. CorX Labs did not run these evaluations. Most are self-reported by the lab that built the model, which means they were produced under that lab's own choice of prompt, scaffold and number of attempts — so treat them as a starting point for a shortlist, not as a settled ranking.
A score someone other than the model's maker measured is marked Independent and names its measurer. Those are the stronger numbers on this page — an outside harness has no reason to flatter anyone — and there are not many of them.
Where a figure has not been published, the cell reads Not reported rather than an estimate. Nothing here is inferred, interpolated or guessed. Each model records the month its row was last checked. Full method and caveats.
Questions
DeepSeek-V4-Pro or DeepSeek-V3.2?
Is DeepSeek-V4-Pro better than DeepSeek-V3.2?
These two models have no benchmark in common with published figures for both, so there is nothing to compare directly. The specification and price rows below are still like for like.
Which is cheaper, DeepSeek-V4-Pro or DeepSeek-V3.2?
At least one of these does not have a published per-token price, so cost cannot be compared like for like. The open-weight model can be run on your own hardware instead.
What is the difference between DeepSeek-V4-Pro and DeepSeek-V3.2?
DeepSeek-V4-Pro takes 1M tokens of context against 164K — 6.4× more room for long documents or a large codebase.
