DeepSeek V3
DeepSeek · released Dec 25, 2024 · deepseek-ai/DeepSeek-V3
- Type
- Open weights
- Params
- 685B
- Context
- 164K
37B active per word · about 123K words of context
Our take
Written Aug 1, 2026Built for coding and reasoning workloads, this large downloadable model from DeepSeek uses a mixture-of-experts design that keeps only 37 billion parameters active per token out of 684.5 billion total. It offers a wide context window and strong benchmarked coding performance, though its licence terms remain unverified and it handles text only.
Pick this for coding workloads where benchmarked scores matter, or cost-sensitive high-volume input. Choose it when throughput is critical, or when downloadable weights matter but you can verify licence terms yourself. Skip it if you need multimodal input, require a verified permissive licence, or prioritise creative writing over coding.
The case for it
- Strongest measured skill is coding, with a 29.1-point gap over its own overall text score.
- Dramatic parameter efficiency: only 37 billion active per token from 684.5 billion total.
- Wide context window of 163,840 tokens for long-document tasks.
- Lowest input cost among its own tracked offers, with a 3.46× spread between cheapest and most expensive.
The case against it
- Licence terms are unverified for commercial safety; not disclosed in our data.
- Text-to-text only; no image, video, or audio handling.
- Creative writing and instruction following lag its own coding score by 38.8 and 43.9 points.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)98th of 143 · 1358.5
CodingWriting and fixing code on its own
Arena Coding104th of 143 · 1387.7
AgenticPlanning, calling tools, staying on task
Nobody we watch has scored DeepSeek V3 for this. We would take the rating from Arena Agent (IPS).
WritingWe do not rate this
Two boards come close and neither tests writing: Arena Creative Writing asks people which of two replies they prefer, and LiveBench Language tests whether a model understood a passage. So we show where DeepSeek V3 placed and give it no mark out of five.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.
Every published score for this model7 scoresEvery figure we hold, from 7 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Radeon RX 7900 XT · 20 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Memory use by level
Against a 24 GB card.
Check against your own machine → · All 71 devices, with every size →
Or rent it from someone else
Cheapest of 6 live listings. Picked at the widest standard context we hold, within one quantisation slice, so the numbers beside it are a price one host actually charges.
- per 1M tokens
- $0.26 in / $1.03 out
- Context served
- 164K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| DeepInfrafp4 | $0.32 / $0.89 | 164K | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp4 | $0.32 / $0.89 | 164K | 14 tok/s | No | No | Confirmed |
| Novita AI | $0.89 / $0.89 | 64K | not measured | Unknown | Unknown | Unknown |
| OpenRouter | $0.26 / $1.03 | 164K | not measured | Unknown | Unknown | Unknown |
| StreamLake | $0.26 / $1.03 | 128K | 29 tok/s | No | Yesunknown period | Unknown |
| Novita AIfp8 | $0.40 / $1.30 | 64K | 23 tok/s | No | No | Confirmed |
DeepInfra
Novita AIAcross the 6 listings we hold: 3 say they do not train on prompts, 0 say they do and 3 do not say. 2 appear in the zero-retention registry we check; the rest are unknown to us rather than confirmed either way.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
- ✓
- Supported
- ✗
- Not supported
- Not published
- host gave no parameter list
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| DeepInfrafp4 | |||
| DeepInfrafp4 | ✓ | ✓ | ✓ |
| Novita AI | |||
| OpenRouter | ✓ | ✓ | ✓ |
| StreamLake | ✓ | ✓ | ✓ |
| Novita AIfp8 | ✓ | ✗ | ✗ |
Tool calling: 4 of 6 listings say yes, 2 publish no parameter list. JSON output: 3 of 6 listings say yes, 1 says no, 2 publish no parameter list. Strict schema: 3 of 6 listings say yes, 1 says no, 2 publish no parameter list.
Models people weigh against DeepSeek V3
When we formed this view
Dates behind this page
Prices last checked 5h ago
What we do not know about this model yet
- 2 of 6 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 3 of 6 listings do not say whether they train on prompts.
- We hold no cached-input rate for any of its listings.
Licence and identifiers
What the licence allowsWe hold no licence record for this model. Inside are the identifiers you need to pull it — its Hugging Face repo where we have one, our slug and a machine-readable card.
Licence
We hold no licence record for this model yet.
Identifiers
- Hugging Face
- deepseek-ai/DeepSeek-V3
- Architecture
- Mixture of experts
- Modality record
- text->text
- Catalogue slug
- deepseek-deepseek-v3