DeepInfra Inference API & Custom LLM GPU Deployments

Offering overview

Price aggressiveness sets this inference provider apart. Its OpenAI-compatible API hosts a long list of models including Llama 3.x and 4, DeepSeek V3/V4/R1, Qwen 2.5 through 3.8, Gemma 3 and 4, Mistral and Nemo, Phi-4, NVIDIA Nemotron, Kimi, plus proprietary Gemini and Claude access, FLUX image generation, Voxtral speech-to-text and a range of embedding models. Three service tiers (Flex, Standard, Priority) trade latency against cost. Teams that need their own weights can rent A100, H100, H200, B200 or B300 GPUs billed by the minute with autoscaling, or arrange dedicated DGX clusters through sales. The platform is SOC 2 and ISO 27001 certified, supports spending limits, and allows up to 200 concurrent requests per account by default. Startups and cost-sensitive production workloads are the sweet spot. Mistral Nemo starts at USD 0.019 per 1M input tokens, Llama 3.1 8B at USD 0.02, embeddings at USD 0.005, and an on-demand A100 80GB at USD 0.89 per GPU-hour.

Technical specifications

Containers

Unknown

Kubernetes

Unknown

Root access

Unknown

API access

Unknown

Autoscaling

Unknown

Persistent processes

Unknown

AI capabilities

Model serving

Llama 3.1, Llama 3.3 70B, Llama 4 Scout, DeepSeek-V3, DeepSeek-V4, DeepSeek-R1, Qwen2.5 72B, Qwen3, Qwen3.5, Qwen3.6, Qwen3-Coder, Gemma 3, Gemma 4, Mistral Small, Mistral Nemo, Phi-4, NVIDIA Nemotron, Kimi K2, Gemini, Claude, FLUX, Voxtral, bge-m3

AI frameworks

Not stated

Inference endpoints

Unknown

Fine-tuning

Unknown

Vector database

Unknown

Model catalog

Unknown

Monitoring

Not stated

Evaluation

Not stated

Security & compliance

Geography

Provider headquarters are separate from offering infrastructure and availability.

Available plans

Pricing can change. Promotional prices are shown separately from standard prices.

Questions & answers

What does DeepInfra's API support?

DeepInfra offers an OpenAI-compatible API for open and proprietary models, including Llama, DeepSeek, Qwen, Gemini and Claude. It also supports FLUX image generation, Voxtral speech-to-text and embedding models.

Who is DeepInfra for?

DeepInfra serves startups and teams with cost-sensitive production workloads, as well as teams that need GPU deployments for their own model weights.

Can teams deploy custom LLMs on DeepInfra?

Teams can rent A100, H100, H200, B200 or B300 GPUs billed by the minute with autoscaling. Dedicated DGX clusters can be arranged through sales.

What API service tiers does DeepInfra offer?

DeepInfra offers Flex, Standard and Priority tiers that trade latency against cost.

Does DeepInfra support spending limits and concurrent requests?

The platform supports spending limits and allows up to 200 concurrent requests per account by default.