See how ChatGPT, Perplexity and Google AI Overviews describe you today
Free AI Visibility AuditDeepSeek V3.2 Exp is an experimental text model from DeepSeek, released as an intermediate step between V3.1 and the architectures the lab plans next, and built around a fine-grained sparse attention mechanism called DeepSeek Sparse Attention.
DeepSeekVerified 24/09/2026Released 29/09/202501 / decision
Each figure sits next to the middle of the field, so it can be read as dear or cheap, wide or narrow, rather than floating on its own. Compared against the 126 models we track, not against an absolute standard.
Capability
Not measured
No published benchmark scores yet
Input / 1M tokens
$0.27
field median $0.30
Output / 1M tokens
$0.41
field median $1.25
Context
164K
field median 524K
02 / overview
Four questions, answered separately, because somebody arrives at one of them rather than at the top of the page.
The job it was built for, and the job it is not for.
DeepSeek V3.2 Exp is a text-only large language model published as a research staging post rather than a finished product line. Its job is to put DeepSeek Sparse Attention in front of real workloads and see how it holds up at long input lengths. It is not a multimodal model, and the "Exp" label is a signal from the lab that this is not the build you standardise a production stack on.
Where it sits in its line, and when a sibling is the better pick.
DeepSeek V3.2 Exp first appeared at the end of September 2025, sitting between V3.1 and whatever DeepSeek ships next. It is the right pick when you want to test how the sparse attention approach behaves on your own long-context text work, or when cost per token dominates the decision. It is the wrong pick when you need a stable target for a shipped product, or anything involving images, audio or video.
The API, the apps it powers, and what its limits let you do.
DeepSeek V3.2 Exp is reached through the DeepSeek API and the usual third-party inference routers, with open weights available in the same way as the rest of the DeepSeek line. The context window takes a large document set, a long transcript or a sizeable codebase in one pass, and the output ceiling is generous enough for full file rewrites and long structured reports rather than short answers. Everything in and out is text.
What changes because this exists, or why it does not.
The point of DeepSeek V3.2 Exp is that long-context work gets cheaper to attempt. Jobs that were awkward to justify at scale, running a whole corpus through a model repeatedly, batch-processing long transcripts, evaluating a large set of documents, become easy to run and rerun. For most teams this is a cost and experimentation story rather than a capability story.
Follows DeepSeek V3.1 Terminus. Superseded by DeepSeek V3.2. See the whole line.
03 / evidence
Split by category, so a strong number never hides a thin evidence base. Verified means we read it on the benchmark's own published results; a provider's claim about its own model is shown and labelled rather than dropped.
No published benchmark scores for this model yet.
We publish a score only where we can link the result to where it was published. Until a benchmark result for this model exists in a source we read, this section stays empty rather than being filled with a provider's marketing figure.
How we decide what counts as evidence04 / ledger
Every published row, grouped by category, each compared with the best published score on the same benchmark. 'Is 64% good' is a question nobody can answer; '26 points behind the leader' is one anybody can.
Nothing in the ledger yet.
Each row here carries a score, the benchmark it came from, what the leading model scored on the same test, and a link to the published result. Rows appear as results are published and read.
How we decide what counts as evidence05 / capability
Where this model is strong, and against how many peers. Ranks are against models with evidence in that category, not against everything we track: ranking against models nobody tested would rank who published, not who is better.
No category scores to shape yet.
A category score is the weighted mean of the benchmarks published for it. With no published rows there is nothing to average, and an empty chart drawn at zero would say something false.
How we decide what counts as evidence06 / cost
List API rates as last read from the provider, with the source on every row, plus every change we have recorded since we started tracking it.
| Charge | Price | Unit | Read on | Source |
|---|---|---|---|---|
| Input | $0.27 | per 1M tokens | 2026-09-24 | Check |
| Output | $0.41 | per 1M tokens | 2026-09-24 | Check |
DeepSeek V3.2 Exp is among the cheapest models you can call for work of this size, with output priced close enough to input that long generations do not blow up a budget. Compared with the frontier models from the large US labs, the gap is not marginal, it is the kind of difference that changes which internal projects are worth running at all.
| Workload | Input / month | Output / month | Cost |
|---|---|---|---|
| A small product team | 20M tokens | 5M tokens | $7.45 |
| A busy support assistant | 200M tokens | 40M tokens | $70.40 |
| A document pipeline | 1000M tokens | 100M tokens | $311.00 |
List API rates, no caching and no batch discount, which both providers offer and which change the answer a great deal. Treat these as the ceiling, not the bill.
07 / specs
As listed by the provider's own catalogue and re-read every few hours. Anything absent is absent there too.
| Context window | 163,840 tokens |
|---|---|
| Maximum output | 65,536 tokens |
| Modalities | text |
| Released | 29/09/2025 |
| Status | Current |
| Catalogue identifier | deepseek/deepseek-v3.2-exp |
08 / lineage
What this model replaced, what replaced it, and what else its provider has in the field.
Also from DeepSeek
09 / line
Every model in this family in release order, so a page from eight months ago says in one glance that two newer ones exist.
Came before
← DeepSeek V3.1 TerminusCame after
DeepSeek V3.2 →Ordered by release date and worked out from the naming, so a new member slots in as soon as its page exists. A retired model keeps its page and its place in the line.
10 / notes
What this model changes for a brand trying to be cited in AI answers, and every change we have logged since it launched.
What it changes for you
A model this cheap to run at long context means more products can afford to read a lot before they answer, so the volume of AI-generated answers citing sources goes up, not down. For a SaaS brand, that rewards documentation and comparison content that survives being read in bulk alongside competitors rather than in isolation. It also means your category is increasingly being summarised by models you have no commercial relationship with, embedded in tools you have never heard of.
Buyers meet DeepSeek V3.2 Exp mainly through the API and through inference providers hosting the open weights, rather than in a mass-market consumer app. In practice it turns up inside other people's products: chat features, coding assistants and document tools that chose it on cost, often without naming it in the interface. Its consumer surface is the DeepSeek assistant itself, which is where most non-developer exposure to the DeepSeek name comes from.
Nothing published here yet. Changes appear within hours of a provider announcing them.
11 / questions
The things people ask about this model, answered from what is on this page rather than from anywhere else.
Models change what gets cited. We measure whether AI answers name your brand or your competitors across every assistant, and show you what to change.