The AI arena is free today

Open Playground
API Provider15 active models13 organizations

DeepInfra: API pricing, speed & models

DeepInfra hosts 15 active AI models, with input pricing from $0.06 per 1M tokens, with median throughput of 83 characters/sec, and P95 time to first token of 9.98s, with 96.9% success rate over 7 days. Compare DeepInfra's API speed, pricing, and reliability against other inference providers.

Median throughput83 c/s
Median TTFT1.53s
P95 TTFT9.98s
Success rate (7d)96.9%
From$0.06 /M tok

Catalog

Moonshot AI6Alibaba Cloud / Qwen Team6Xiaomi5ByteDance3DeepSeek3Google2NVIDIA1Zhipu AI1
Type
Price
27 models
Model
Qwen3 32B26.95 QPS
Qwen3 30B A3B82.57 QPS
At a glance

DeepInfrapricing, performance & catalog

The citable facts about DeepInfra's 15 models — sourced from provider APIs and refreshed continuously.

Lowest price
Nemotron 3 Nano (30B A3B) at $0.060 per 1M input tokens
Highest median throughput
Nemotron 3 Nano (30B A3B) at 784 chars/s
Lowest median TTFT
MiMo-V2.5-Pro at 0.36s
Largest context
DeepSeek-V4-Flash-0731 at 1.0M tokens
Catalog
15 active models from 13 organizations

FAQ

Common questions about DeepInfra.

What is DeepInfra?

DeepInfra is an API provider that hosts large language models. Active models: 15; From (input): $0.06 / 1M tok; Median throughput: 83 c/s; Median TTFT: 1.53s; Success rate (7d): 96.9%.

How many models does DeepInfra offer?

DeepInfra currently serves 15 active models out of 54 historical offerings on LLM Stats.

What is DeepInfra's API pricing?

DeepInfra input pricing starts from $0.06 per 1M tokens, with the most expensive offering at $1.74 per 1M tokens. See the Pricing tab above for the full per-model breakdown.

How fast is DeepInfra?

DeepInfra delivers a median throughput of 83 characters per second and P95 time to first token of 9.98s across its model catalog. See the Models tab for per-model throughput and TTFT breakdowns.

Is DeepInfra reliable?

DeepInfra has a 96.9% success rate across 747 API calls in the last 7 days, with a 3.1% error rate.

Does DeepInfra support function calling?

Yes. 27 of 15 models on DeepInfra support function calling (tool use). The Capabilities tab lists which specific models accept tool definitions.

Does DeepInfra support JSON mode and structured output?

Yes. 27 of 15 models on DeepInfra support structured output (JSON mode / schema-constrained generation). The Capabilities tab shows which specific models accept response_format or json_schema parameters.

Does DeepInfra offer batch inference?

Yes. 24 of 15 models on DeepInfra support batch inference for cheaper, asynchronous workloads.

Does DeepInfra support multimodal models?

Yes. DeepInfra's catalog includes 7 vision-capable models. See the Models and Capabilities tabs for the full per-model breakdown.

Whose models does DeepInfra host?

DeepInfra hosts models from ByteDance, DeepSeek, Google, Moonshot AI, NVIDIA, and Alibaba Cloud / Qwen Team, plus 7 more. See the Models tab for the full catalog grouped by creator.

How do I start using DeepInfra?

Sign up at https://deepinfra.com/ to get an API key, then call DeepInfra's API directly from your application. Most clients work out of the box by pointing the OpenAI SDK at DeepInfra's base URL with your key. Use the Models and Pricing tabs above to pick the right model for your latency, cost, and context-window requirements.