See how ChatGPT, Perplexity and Google AI Overviews describe you today
Free AI Visibility AuditDeepSeek V3 0324 is a text-only general chat model from DeepSeek, a mixture-of-experts system of 685B parameters and the March 2025 refresh of the flagship DeepSeek V3 family.
DeepSeekVerified 06/10/2026Released 24/03/202501 / 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 232 models we track, not against an absolute standard.
Capability
42.0/100
#17 of 22 with published scores
Input / 1M tokens
$0.29
field median $0.39
Output / 1M tokens
$1.14
field median $2.00
Context
164K
field median 500K
Read the coverage before the score. That figure rests on 20% of the weighting: the rest of the categories have no published evidence for this model, and are not counted rather than counted as zero.
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 0324 is a general-purpose chat and instruction model, built to handle the everyday work of drafting, answering, summarising and coding against long inputs. It is the direct successor to the earlier DeepSeek V3 release rather than a new line. It handles text only, so anything involving images, audio or document scans needs a different model in the chain.
Where it sits in its line, and when a sibling is the better pick.
DeepSeek V3 0324 was first seen on 24 March 2025, arriving as an update to the existing DeepSeek V3 chat model rather than a separate product. It is the right pick when you want the current version of the V3 line for high-volume text work at low cost. It is the wrong pick if your workload involves any non-text input, or if you need a vendor-hosted consumer assistant rather than a model you call yourself.
The API, the apps it powers, and what its limits let you do.
DeepSeek V3 0324 is reached through the DeepSeek API and through the aggregators that resell it, and it also sits behind the DeepSeek chat assistant. The context window runs to roughly 164k tokens with an unusually high output ceiling, so a single call can take in a long codebase, a contract set or a batch of support tickets and return a full document rather than a fragment. Everything in and out is plain text.
What changes because this exists, or why it does not.
DeepSeek V3 0324 makes bulk text work cheap enough to stop rationing it, so jobs that were previously batched or sampled, classifying every ticket, rewriting every product page, reviewing every pull request, can run across the whole set. The large output allowance also removes the usual stitching work, where a long deliverable had to be generated in pieces and reassembled.
Superseded by DeepSeek V3.1. 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.
Published rows
1
of 19 benchmarks we track
Independently verified
0
1 from the provider
Weighting covered
20%
the rest is not measured, not zero
Agentic
0/3
Not measured
Coding
1/3
Provider claim
Reasoning
0/3
Not measured
Multimodal
0/3
Not measured
Knowledge
0/2
Not measured
Multilingual
0/2
Not measured
Instruction following
0/1
Not measured
Maths
0/2
Not measured
04 / 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.
| Benchmark | Score | Best published | Weight | Source |
|---|---|---|---|---|
| SWE-bench VerifiedReal GitHub issues, resolved and tested. | 42% | MiniMax M2.575.8% · 33.8 behind | 40% | SWE-bench Verified · SWE-ExpProvider claim |
Benchmark data from SWE-bench. Every row links the published result it was read from.
05 / 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.
| Category | Score | Rank | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|
| Agentic | Not measured | Not ranked | 22% | 0 of 3 | Not measured |
| Coding | 42.0 | #17 of 2224th percentile | 20% | 1 of 3 | Provider claim |
| Reasoning | Not measured | Not ranked | 17% | 0 of 3 | Not measured |
| Multimodal | Not measured | Not ranked | 12% | 0 of 3 | Not measured |
| Knowledge | Not measured | Not ranked | 12% | 0 of 2 | Not measured |
| Multilingual | Not measured | Not ranked | 7% | 0 of 2 | Not measured |
| Instruction following | Not measured | Not ranked | 5% | 0 of 1 | Not measured |
| Maths | Not measured | Not ranked | 5% | 0 of 2 | Not measured |
Category weights are ours and they are a judgement: weighted for what a SaaS team buys a model to do. Agentic and coding lead because that is where the work is. Published here rather than hidden, and set out in full on the methodology page.
06 / 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 |
|---|---|---|---|---|
| Output | $1.14 | per 1M tokens | 2026-10-05 | Check |
| Input | $0.29 | per 1M tokens | 2026-10-05 | Check |
DeepSeek V3 0324 sits at the low end of the market, cheap enough that input volume stops being the thing you design around, with output charged at a few times the input rate as is standard. Against the frontier models from the larger US labs it is an order of magnitude less expensive, which is the main reason teams put it on high-throughput internal jobs.
| Workload | Input / month | Output / month | Cost |
|---|---|---|---|
| A small product team | 20M tokens | 5M tokens | $11.50 |
| A busy support assistant | 200M tokens | 40M tokens | $103.60 |
| A document pipeline | 1000M tokens | 100M tokens | $404.00 |
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.
Every reading we have taken, kept whether it changed or not. This is the part nobody can copy in a week.
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 | 115,200 tokens |
| Modalities | text |
| Released | 24/03/2025 |
| Status | Current |
| Catalogue identifier | deepseek/deepseek-chat-v3-0324 |
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
The first we track in this line.
Came after
DeepSeek V3.1 →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
Cheap, capable general models like DeepSeek V3 0324 mean more of your competitors can afford to generate content and run retrieval at scale, so volume stops being a differentiator. What still decides whether a SaaS brand gets named in an AI answer is whether the underlying facts about your product are clear, consistent and available to be retrieved. Worth noting that this model is text-only, so anything you communicate through images, video or slides is invisible to it.
A buyer meets DeepSeek V3 0324 most often as a developer, through the DeepSeek API or a routing layer that lists it among cheap general models. Consumers meet the V3 line through the DeepSeek assistant app and site. It is not the engine behind the mainstream Western AI search surfaces, so it shapes what gets built more than what gets cited.
Model: DeepSeek V3 0324 Input was $0.29 per 1M tokens, now $0.25 per 1M tokens Output was $1.14 per 1M tokens, now $1 per 1M tokens
Model: DeepSeek V3 0324 Input was $0.25 per 1M tokens, now $0.29 per 1M tokens Output was $1 per 1M tokens, now $1.14 per 1M tokens
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.