# DeepSeek vs Nebius

> Nebius serves DeepSeek and 60+ other open models with EU or US placement. DeepSeek's own API is cheap but stores data in China. Residency is usually the deciding factor.

Canonical: https://www.subconscious.dev/compare/deepseek-vs-nebius · By The Subconscious Team · Updated September 30, 2026

## How they compare

DeepSeek's weights are open, and Nebius's Token Factory is one of the hosts that serves them, alongside Llama, Qwen, GLM, Kimi and GPT-OSS. So the matchup is mostly about the wrapper. DeepSeek's first-party API is cheap, with V4 Pro at $1.32 in and $3.96 out at peak and half that off-peak, but hosted data is stored in China. Nebius, based in Amsterdam, offers dedicated endpoints with a 99.9% SLA and optional EU or US placement, prices starting at $0.06 per million input tokens, and serving of uploaded fine-tunes at the same token pricing.

For a European enterprise, or anyone who cannot send data to China, Nebius is the practical way to run DeepSeek-class models. It also sells raw GPUs, from H100s at $2.15 an hour preemptible, so a team can move from tokens into training on one account. DeepSeek direct suits cost-first teams that accept its data terms and want the newest DeepSeek releases from the source. Watch the churn, though: DeepSeek retires and reprices models often. Nebius requires a $25 minimum first payment and has no free trial.

## What each one does

### DeepSeek

DeepSeek is the Chinese lab whose open-weight models reset price expectations for the whole market. Its API now serves two models, both with 1M context and 384K max output. V4.1 Flash shipped September 10, 2026 with built-in image understanding at $0.30 in and $1.20 out at peak. V4 Pro, generally available since August 13, costs $1.32 in and $3.96 out at peak. Cache hits cost a few cents per million or less, and the weights ship on Hugging Face under an MIT license.

### Nebius

Nebius is an Amsterdam-headquartered AI cloud and the strongest European alternative to the US hyperscalers. It sells raw NVIDIA GPU compute, from H100s at $2.15 an hour preemptible up to GB300 NVL72 racks, and it has begun adding Vera Rubin. Hyperscale buyers back it: a Microsoft capacity deal worth about $17.4B in September 2025, then a Meta agreement worth up to about $27B in March 2026.

## Which is best, and when

### Choose DeepSeek for

- Cost-first teams comfortable with China data storage
- Off-peak batch at half price
- Getting new DeepSeek releases from the source

### Choose Nebius for

- Running DeepSeek-class models with EU placement
- Serving fine-tuned DeepSeek checkpoints with an SLA
- Growing from inference into GPU training

## At a glance

| Attribute | DeepSeek | Nebius |
|---|---|---|
| Model access | Open weights (MIT) | Open weights, 60+ models |
| Flagship models | DeepSeek V4.1 Flash, V4 Pro | DeepSeek, Qwen, GLM, Kimi, GPT-OSS |
| Speed | ~35 tok/s on V4 Pro | Among top hosts on throughput |
| Price | Off-peak hours at half price | From $0.06 per 1M input |
| Customization | Open weights to fine-tune | Serve uploaded fine-tunes |
| Deployment | First-party API, Hugging Face weights | Token Factory, dedicated, raw GPUs |
| Long context | 1M, 384K max output | Varies by model |

## FAQ

### What is the difference between DeepSeek and Nebius?

Nebius serves DeepSeek and 60+ other open models with EU or US placement. DeepSeek's own API is cheap but stores data in China. Residency is usually the deciding factor.

### When should I choose DeepSeek over Nebius?

Cost-first teams comfortable with China data storage; Off-peak batch at half price; Getting new DeepSeek releases from the source.

### When should I choose Nebius over DeepSeek?

Running DeepSeek-class models with EU placement; Serving fine-tuned DeepSeek checkpoints with an SLA; Growing from inference into GPU training.

### Is DeepSeek or Nebius cheaper?

DeepSeek: Off-peak hours at half price. Nebius: From $0.06 per 1M input. The cheaper choice depends on the model and workload.

### Which has more context, DeepSeek or Nebius?

DeepSeek: 1M, 384K max output. Nebius: Varies by model.

## Running long-horizon agents?

If your agents run past 200K tokens, compare both against Subconscious: [Subconscious vs DeepSeek](https://www.subconscious.dev/compare/subconscious-vs-deepseek.md), [Subconscious vs Nebius](https://www.subconscious.dev/compare/subconscious-vs-nebius.md).

Full profiles: [DeepSeek](https://www.subconscious.dev/providers/deepseek.md), [Nebius](https://www.subconscious.dev/providers/nebius.md).