Nebius: Models Intelligence, Performance & Price
Analysis of Nebius's models across key metrics including quality, price, output speed, latency, context window & more. This analysis is intended to support you in choosing the best model provided by Nebius for your use-case. For more details including relating to our methodology, see our FAQs. Models analyzed: Llama 3.3 70B Fast, Llama 3.3 70B Base, Llama 3.1 405B Base, Mistral NeMo Base, DeepSeek R1 Distill Llama 70B Base, DeepSeek V3 0324 (Mar '25) Fast, DeepSeek V3 0324 (Mar '25), DeepSeek R1 0528 (May '25), Phi-4, Llama Nemotron Ultra Reasoning Base, Qwen3 4B (Reasoning) Fast, QwQ-32B Fast, Qwen3 235B (Reasoning) Base, Qwen3 32B (Reasoning) Base, Qwen3 30B A3B (Reasoning) Fast, Qwen3 30B A3B (Reasoning) Base, Qwen3 14B (Reasoning) Base, Llama 3.1 70B Base, Llama 3.1 8B Fast, Llama 3.1 8B Base, Gemma 2 9B Fast, DeepSeek R1 (Jan '25) Base, DeepSeek R1 (Jan '25) Fast, DeepSeek V3 (Dec '24), Qwen2.5 72B, Qwen2.5 72B Fast, Qwen2.5 Instruct 32B Fast, and Qwen2.5 Instruct 32B Base.
Link:
Nebius Model Comparison Summary
Intelligence:
DeepSeek R1 0528 (May '25) and 
Qwen3 235B (Reasoning) Base are the highest quality models offered by Nebius, followed by 
Llama Nemotron Ultra Reasoning Base, DeepSeek R1 (Jan '25) Base & DeepSeek R1 (Jan '25) Fast.Output Speed (tokens/s):
Gemma 2 9B Fast (193 t/s) and 
Llama 3.1 8B Fast (184 t/s) are the fastest models offered by Nebius, followed by Qwen3 4B (Reasoning) Fast, Llama 3.3 70B Fast & Qwen3 30B A3B (Reasoning) Fast.Latency (seconds): Llama 3.1 8B Fast (0.47s) and Qwen3 4B (Reasoning) Fast (0.49s) are the lowest latency models offered by Nebius, followed by Gemma 2 9B Fast, Qwen3 30B A3B (Reasoning) Base & QwQ-32B Fast.Blended Price ($/M tokens): Llama 3.1 8B Base ($0.03) and Llama 3.1 8B Fast ($0.04) are the cheapest models offered by Nebius, followed by Gemma 2 9B Fast, 
Mistral NeMo Base & Qwen2.5 Instruct 32B Base.Context Window Size: DeepSeek R1 0528 (May '25) (164k) and Llama Nemotron Ultra Reasoning Base (131k) are the largest context window models offered by Nebius, followed by QwQ-32B Fast, Qwen2.5 72B & Qwen2.5 72B Fast.
DeepSeek R1 0528 (May '25) and Qwen3 235B (Reasoning) Base are the highest quality models offered by Nebius, followed by Llama Nemotron Ultra Reasoning Base, DeepSeek R1 (Jan '25) Base & DeepSeek R1 (Jan '25) Fast.Output Speed (tokens/s): Gemma 2 9B Fast (193 t/s) and Llama 3.1 8B Fast (184 t/s) are the fastest models offered by Nebius, followed by Qwen3 4B (Reasoning) Fast, Llama 3.3 70B Fast & Qwen3 30B A3B (Reasoning) Fast.Latency (seconds): Llama 3.1 8B Fast (0.47s) and Qwen3 4B (Reasoning) Fast (0.49s) are the lowest latency models offered by Nebius, followed by Gemma 2 9B Fast, Qwen3 30B A3B (Reasoning) Base & QwQ-32B Fast.Blended Price ($/M tokens): Llama 3.1 8B Base ($0.03) and Llama 3.1 8B Fast ($0.04) are the cheapest models offered by Nebius, followed by Gemma 2 9B Fast, Mistral NeMo Base & Qwen2.5 Instruct 32B Base.Context Window Size: DeepSeek R1 0528 (May '25) (164k) and Llama Nemotron Ultra Reasoning Base (131k) are the largest context window models offered by Nebius, followed by QwQ-32B Fast, Qwen2.5 72B & Qwen2.5 72B Fast.Highlights
Intelligence
Artificial Analysis Intelligence Index; Higher is better
Speed
Output Tokens per Second; Higher is better
Price
USD per 1M Tokens; Lower is better
Parallel Queries:
Prompt Length:
Features | Model Intelligence | Price | Output tokens/s | Latency | End-to-End Response Time | ||||
|---|---|---|---|---|---|---|---|---|---|
Further Analysis | |||||||||
 | DeepSeek R1 0528 (May '25) | 164k | 68 | $1.20 | 30.5 | 0.66 | 82.57 | 65.53 | |
| Qwen3 235B (Reasoning) Base | 33k | 62 | $0.30 | 29.6 | 0.61 | 85.21 | 67.68 | |
| Llama Nemotron Ultra Reasoning Base | 131k | 61 | $0.90 | 42.0 | 0.64 | 60.20 | 47.65 | |
| DeepSeek R1 (Jan '25) Base | 128k | 60 | $1.20 | 32.5 | 0.65 | 88.19 | 72.17 | |
| DeepSeek R1 (Jan '25) Fast | 128k | 60 | $3.00 | 79.2 | 0.67 | 36.62 | 29.63 | |
| Qwen3 32B (Reasoning) Base | 33k | 59 | $0.15 | 45.1 | 0.58 | 56.05 | 44.37 | |
| QwQ-32B Fast | 131k | 58 | $0.75 | 75.1 | 0.52 | 40.32 | 33.15 | |
| Qwen3 14B (Reasoning) Base | 33k | 56 | $0.12 | 87.2 | 0.52 | 29.20 | 22.94 | |
| Qwen3 30B A3B (Reasoning) Fast | 33k | 56 | $0.45 | 137.7 | 0.54 | 18.70 | 14.53 | |
| Qwen3 30B A3B (Reasoning) Base | 33k | 56 | $0.15 | 116.8 | 0.50 | 21.90 | 17.12 | |
| DeepSeek V3 0324 (Mar '25) Fast | 128k | 53 | $3.00 | 87.5 | 0.68 | 6.40 | N/A | |
| DeepSeek V3 0324 (Mar '25) | 128k | 53 | $0.75 | 27.7 | 0.67 | 18.71 | N/A | |
| DeepSeek R1 Distill Llama 70B Base | 128k | 48 | $0.38 | 59.2 | 0.59 | 42.81 | 33.78 | |
| Qwen3 4B (Reasoning) Fast | 33k | 47 | $0.12 | 156.5 | 0.49 | 16.47 | 12.78 | |
| DeepSeek V3 (Dec '24) | 128k | 46 | $0.75 | 23.8 | 0.65 | 21.62 | N/A | |
| Llama 3.3 70B Fast | 128k | 41 | $0.38 | 153.8 | 0.52 | 3.77 | N/A | |
| Llama 3.3 70B Base | 128k | 41 | $0.20 | 37.1 | 0.66 | 14.15 | N/A | |
| Llama 3.1 405B Base | 128k | 40 | $1.50 | 31.8 | 0.69 | 16.41 | N/A | |
| Qwen2.5 72B | 131k | 40 | $0.20 | 24.8 | 0.70 | 20.83 | N/A | |
| Qwen2.5 72B Fast | 131k | 40 | $0.38 | 66.9 | 0.56 | 8.03 | N/A | |
|  Phi-4 | 16k | 40 | $0.15 | 110.1 | 0.52 | 5.06 | N/A | |
| Qwen2.5 Instruct 32B Fast | 128k | 37 | $0.20 | 86.2 | 0.53 | 6.33 | N/A | |
| Qwen2.5 Instruct 32B Base | 128k | 37 | $0.10 | 59.5 | 0.57 | 8.96 | N/A | |
| Llama 3.1 70B Base | 128k | 35 | $0.20 | 29.8 | 0.66 | 17.45 | N/A | |
| Llama 3.1 8B Fast | 128k | 24 | $0.04 | 183.5 | 0.47 | 3.20 | N/A | |
| Llama 3.1 8B Base | 128k | 24 | $0.03 | 61.3 | 0.55 | 8.71 | N/A | |
| Gemma 2 9B Fast | 8k | 22 | $0.04 | 193.4 | 0.49 | 3.08 | N/A | |
| Mistral NeMo Base | 128k | 20 | $0.06 | 49.6 | 0.56 | 10.65 | N/A | |
Key definitions
Context window: Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).
Output Speed: Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).
Latency (Time to First Token): Time to first token received, in seconds, after API request sent. For reasoning models which share reasoning tokens, this will be the first reasoning token. For models which do not support streaming, this represents time to receive the completion.
Price: Price per token, represented as USD per million Tokens. Price is a blend of Input & Output token prices (3:1 ratio).
Output Price: Price per token generated by the model (received from the API), represented as USD per million Tokens.
Input Price: Price per token included in the request/message sent to the API, represented as USD per million Tokens.
Time period: Metrics are 'live' and are based on the past 72 hours of measurements, measurements are taken 8 times a day for single requests and 2 times per day for parallel requests.