Nebius

Serverless

European AI inference on Token Factory. Two flavors: fast (low latency) and base (cost-efficient). Batch at 50% off.

OpenAI Compatible

Streaming

Batching

Fine-tuning

Embeddings

Vision

Audio

Function Calling

JSON Mode

SOC 2

HIPAA

GDPR

per token

Free tier: 1 dollar in free credits to start

Models 26

API Base https://api.studio.nebius.ai/v1

LLM Inference

Model Catalog (26)

Model	Type	Input $/1M	Output $/1M	Context	Speed
DeepSeek R1 0528 DeepSeek · 671B MoE	llm	$0.800	$2.40	—	—
DeepSeek V3 DeepSeek · 671B MoE	llm	$0.500	$1.50	—	—
DeepSeek V4 Flash DeepSeek 0731	llm	$0.140	$0.280	1.0M	—
DeepSeek V4 Pro DeepSeek	llm	$1.75	$3.50	1.0M	—
DeepSeek V4 Pro DeepSeek 0813	llm	$1.32	$3.96	979k	—
DeepSeek V4.1 Flash DeepSeek	llm	$0.300	$1.20	1.0M	—
GLM 4.5 Zhipu AI · 106B MoE (12B active)	llm	$0.600	$2.20	—	—
GLM 5.1 Z.ai · 744B MoE (40B active)	llm	$1.40	$4.40	203k	—
GLM 5.2 Z.ai	llm	$1.40	$4.40	1.0M	—
GLM 5.3 Z.ai	llm	$1.40	$4.40	1.0M	—
GLM 5.3 Flash Z.ai	llm	$0.150	$0.500	1.0M	—
GPT OSS 120B OpenAI · 117B MoE (5.1B active)	llm	$0.150	$0.600	—	—
Gemma 3 27B Google · 27B	llm	$0.100	$0.300	110k	—
Kimi K2 Moonshot AI · 1T MoE (32B active)	llm	$0.500	$2.40	—	—
Kimi K2.6 Moonshot AI · 1T MoE (32B active)	llm	$0.950	$4.00	262k	60 t/s
Kimi K2.7 Code Moonshot AI · 1T MoE (32B active)	code	$0.950	$4.00	262k	231.26 t/s
Kimi K3 Moonshot AI	llm	$3.00	$15.00	1.0M	—
Llama 3.3 70B Meta · 70B	llm	$0.130	$0.400	—	—
MiniMax M3 MiniMax	llm	$0.300	$1.20	1.0M	—
Nemotron 3.5 Lightning NVIDIA	llm	$0.060	$0.240	1.0M	—
Qwen 3 235B Alibaba · 235B MoE	llm	$0.200	$0.600	—	—
Qwen 3 32B Alibaba · 32B	llm	$0.100	$0.300	—	—
Qwen3 30B A3B Alibaba	llm	$0.100	$0.300	262k	—
Qwen3 Embedding 8B Alibaba · 8B	embedding	$0.010	$0.0000	41k	—
Qwen3.5 397B A17B Alibaba	llm	$0.600	$3.60	262k	—
Qwen3.8 27B Alibaba	llm	$0.450	$3.00	262k	—

Yes, Nebius has a free tier: 1 dollar in free credits to start. Paid usage is billed per model.

Qwen3 Embedding 8B at $0.01 in / $0 out per 1M tokens, based on the 26 models we track for Nebius.

We track 26 Nebius models: 24 LLM, 1 coding, 1 embedding.

Yes. Nebius exposes an OpenAI-compatible endpoint, so you can use the OpenAI SDKs by changing the base URL and API key.