DeepInfra
Cost-effective inference for open-source models
Cost efficiencyOpen modelsScalingReliabilityOpenAI-compatible
About DeepInfra
DeepInfra offers affordable inference for popular open-source models with automatic scaling and high availability. They specialize in making large models accessible at low cost, with a straightforward per-token pricing model.
Key Models
Llama 4 MaverickQwen 3.5Gemma 4
Live Performance Data
DeepInfra Models on Antbase
| Model | Score | Requests | Success Rate | Avg Latency | p95 Latency | Capabilities |
|---|---|---|---|---|---|---|
| Claude Opus 5 | 55.0 | 1 | 10000.0% | 4959ms | 4959ms | |
| Gemini 2.5 Pro | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| All MiniLM L6 V2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Gemma 3 12b It | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Kimi K2.5 Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| DeepSeek R1 0528 Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🧠 |
| Qwen3 235B A22B Instruct 2507 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Gemini 2.5 Flash | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| PaddleOCR VL 0.9B | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 14B | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Qwen3 Next 80B A3B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| GLM 4.6V | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| DeepSeek R1 0528 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🧠 |
| Text2vec Base Chinese | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Fibo Edit | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen2.5 72B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| DeepSeek V3.1 Terminus | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| NVIDIA Nemotron Nano 9B V2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Clip ViT B 32 Multilingual V1 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Nemotron 3 Nano 30B A3B | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Kimi K2.5 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 Coder 480B A35B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Qwen3 VL 30B A3B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Llama 3.2 11B Vision Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 👁🔧 |
| DeepSeek V3 0324 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Llama Guard 4 12B | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Expand | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Remove Background | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Llama 3.1 Nemotron 70B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| GLM 4.6 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 VL 235B A22B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Crystal | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Kimi K2 Thinking | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🧠 |
| GLM 5 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Meta Llama 3.1 70B Instruct Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Gpt Oss 120b Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| MiniMax M2.5 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Claude 3 7 Sonnet Latest | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Gemma 3 4b It | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Janus Pro 1B | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| GLM 4.7 Flash | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 Coder 480B A35B Instruct Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Meta Llama 3 8B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Qwen3 32B | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Qwen3 30B A3B | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Seedream 4.5 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Seed 2.0 Mini | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Kimi K2 Instruct 0905 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| L3.1 70B Euryale V2.2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen Image Max | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Qwen2.5 VL 32B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 👁🔧 |
| L3.3 70B Euryale V2.3 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| NVIDIA Nemotron Nano 12B V2 VL | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| MiniMax M2.1 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Claude 4 Opus | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| All Mpnet Base V2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Mistral Small 24B Instruct 2501 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Gpt Oss 120b | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Bria 3.2 Vector | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Clip ViT B 32 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Phi 4 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Replace Background | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Gemini 1.5 Flash 8b | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| DeepSeek V3 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Claude 4 Sonnet | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Creative | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| MythoMax L2 13b | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Mixtral 8x7B Instruct V0.1 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| DeepSeek V3.2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| GLM 4.7 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Blur Background | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| P Image | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 235B A22B Thinking 2507 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🧠🔧 |
| OlmOCR 2 7B 1025 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Llama 3.3 70B Instruct Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| L3 8B Lunaris V1 Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 Max | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Seed 1.8 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Llama 3.3 Nemotron Super 49B V1.5 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| P Image Edit | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Gpt Oss 20b | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Enhance | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Paraphrase MiniLM L6 V2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen3 Max Thinking | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🧠🔧 |
| DeepSeek OCR | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Gen Fill | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Janus Pro 7B | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Hermes 3 Llama 3.1 405B | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Llama 4 Scout 17B 16E Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Multi Qa Mpnet Base Dot V1 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Fibo | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Meta Llama 3.1 70B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Erase Foreground | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| All MiniLM L12 V2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Meta Llama 3.1 8B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Hermes 3 Llama 3.1 70B | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Gemini 1.5 Flash | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Gemma 3 27b It | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Olmo 3.1 32B Instruct | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Erase | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Wan2.6 Image Edit | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Qwen Image Edit | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Bria 3.2 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| DeepSeek R1 Distill Llama 70B | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🧠 |
| Qwen Image Edit Max | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Kimi K2.7 Code | 50.0 | 25 | 10000.0% | 23728ms | 53673ms | |
| Meta Llama 3.1 8B Instruct Turbo | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| DeepSeek V3.1 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Seedream 4 | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Llama 4 Maverick 17B 128E Instruct FP8 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| NVIDIA Nemotron 3 Super 120B A12B | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Mistral Small 3.2 24B Instruct 2506 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Wan2.6 T2I | 50.0 | 0 | 0.0% | 400ms | 1000ms | |
| Mistral Nemo Instruct 2407 | 50.0 | 0 | 0.0% | 400ms | 1000ms | 🔧 |
| Qwen3.5 0.8B | 45.0 | 2 | 10000.0% | 41258ms | 43761ms | 🔧 |
| Claude Fable 5 | 35.0 | 28 | 0.0% | 2297ms | 3484ms | |
| Claude Sonnet 4 6 | 33.0 | 9 | 0.0% | 1071ms | 1525ms |
How to Use DeepInfra on Antbase
All DeepInfra models are available through Antbase's unified OpenAI-compatible API. You can either route directly to a specific DeepInfra model or let ANT's intelligent router automatically select the best model for your task.
python
from openai import OpenAI
client = OpenAI(
base_url="https://antbase.ai/v1",
api_key="your-ant-api-key",
)
# Let ANT auto-route (may use DeepInfra if optimal)
response = client.chat.completions.create(
model="ant:auto",
messages=[{"role": "user", "content": "Hello!"}],
)
# Or route directly to a DeepInfra model
response = client.chat.completions.create(
model="deepinfra/llama-4-maverick",
messages=[{"role": "user", "content": "Hello!"}],
)Why Use DeepInfra Through Antbase?
- •No separate account needed — access DeepInfra models instantly through your Antbase API key.
- •Intelligent routing — ANT automatically selects the best DeepInfra model for each task, or falls over to another provider if DeepInfra is slow or down.
- •Same token prices — Antbase charges zero markup on DeepInfra token prices. You pay exactly what DeepInfra charges.
- •Automatic failover — if DeepInfra has an outage, your requests seamlessly route to an equivalent model on another provider.