DeepInfra
Pay-as-you-go inference for 100+ open and proprietary models, plus GPUs
Overview
DeepInfra is an AI inference cloud with an OpenAI-compatible API for 100+ open models (Llama, DeepSeek, Qwen, Kimi, FLUX) and resold Claude and Gemini, with priority and flex tiers, a 20%-off batch API, zero data retention and hourly dedicated GPUs. LLMs from $0.019 per million tokens; H100 $2.20 an hour.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 2 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 100+ models incl. Claude and Gemini
- Embeddings from $0.005/M
- Spending limits
- 200 concurrent requests per model
- Flex 0.8x
- Standard 1x
- Priority 1.5x
- Up to 50,000 lines per file
- Chat, completions and embeddings
- A100 $0.89, H100 $2.20
- H200 $2.69, B200 $3.69, B300 $4.89
- DGX clusters on request
What you can produce with DeepInfra
- OpenAI-compatible inference API for 100+ models
- Open models plus resold Claude and Gemini
- Priority and flex service tiers
- Batch API at 20% off
- Hourly dedicated GPUs incl. H100 and B300
- Zero data retention
DeepInfra is one of the cheapest and simplest ways to call a wide range of AI models, with transparent per-token and per-GPU pricing and its own US data centres. Its catalogue now includes premium proprietary models, so "cheap" depends on what you call, there is no free tier, and its compliance is narrower than sometimes claimed. A strong default for cost-conscious developers; enterprises should check the trust centre for the certifications they need.
What we found
DeepInfra is an AI inference cloud that serves more than a hundred models through a simple pay-as-you-go API compatible with OpenAI's. Its catalogue spans open models such as Llama, Mistral, DeepSeek, Qwen and Kimi for text, embedding and reranking models, image models such as FLUX and speech models, and it now also resells proprietary models from Anthropic and Google, including Claude and Gemini. Requests can run at standard, priority or cheaper flex service levels, a batch API processes jobs within 24 hours at 20 per cent below real-time prices, and dedicated GPUs can be rented by the hour. It states zero data retention and owns and runs its GPUs across eight US data centres. Its trust centre lists SOC 2 Type 1, ISO 27001 and HIPAA compliance, with GDPR in progress.
Deep Infra Inc. was founded in September 2022 by Nikola Borisov, its chief executive, Georgios Papoutsis and Yessenzhar Kanapin, who previously built the imo messenger, and is reported to be based in Palo Alto. It raised an $8 million seed in 2023, an $18 million Series A led by Felicis and Georges Harik in 2025 and a $107 million Series B co-led by 500 Global and Georges Harik in May 2026, with Nvidia and Samsung Next taking part, about $133 million in total by our sum. It says token volume grew 25 times between its last two rounds.
There is no free tier: a card or prepaid balance is required before use. Prices are per million tokens and vary widely, from $0.019 in for Mistral Nemo and $0.10 in and $0.32 out for Llama 3.3 70B Turbo to $2.85 in and $14.25 out for Kimi-K3 and $25 out for Claude Opus 5; embeddings start at $0.005 per million. Priority requests cost 1.5 times the base price and flex 0.8 times. Dedicated GPUs cost $0.89 an hour for an A100, $2.20 for an H100 and up to $4.89 for a B300, billed by the minute.
We found no independent reviews. DeepInfra suits developers wanting cheap, fast access to many models through one API.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
About this listing
This entry was compiled from publicly available data including DeepInfra's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with DeepInfra unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to DeepInfra directly →
Spotted something out of date? Suggest an update →
Related reading
Alternatives to DeepInfra
More in LLMs & Chat
