Skip to content

Oxen.ai Review

Oxen.ai combines git-like version control for large datasets with pay-as-you-go access to 200+ image, video, audio and language models, fine-tuning and dedicated GPUs. Free tier; teams from $30 a month.

Free plan · from $30/mo · API

Stork's verdict on Oxen.ai

Oxen.ai is a good fit if you want versioned data, model access and fine-tuning in one place; teams that only need inference will find dedicated APIs simpler.

Oxen.ai reviewed by Stork AI · stork.ai/en/oxen-ai

stork
I am a flat drawing.
One flat drawing and one typed sentence.
This drawing was never animated.
A photo, a cartoon, a painting or a 3D character. From 10 cents a minute. Free to try.
Make a photo talkDevelopers: there is an API

What Oxen.ai does

Oxen.ai, from Oxen Labs, Inc., started as Oxen, an open-source (Apache 2.0, written in Rust) version control system for data that is too big for Git: datasets, model weights, images, video, audio and Parquet files. It works like Git, with branches, commits, diffs and merges, through the oxen command line tool, a Python package and an Oxen Server you can self-host. Oxen says its CLI is 20 times faster than git-lfs and scales to repositories with millions of files and terabytes of data. Repositories can live in Oxen's cloud or your own.

On top of that, the hosted platform now adds a model layer: one API and UI for more than 200 models across text, image, video and audio (Kling 4.0, Seedance 2.5, Veo 3.1, FLUX 3, Nano Banana 2, GPT Image 2, DeepSeek and Claude models, among others), with prompts, reference assets and outputs saved into versioned repositories you can tag, search and trace back to the prompt and model behind each result. You can then fine-tune open models (LLMs, vision-language, image and video) on that data with full or LoRA training and deploy the result to a dedicated GPU endpoint. Oxen also runs Oxen Studios for commissioned AI and VFX work and a weekly community paper-reading session.

Oxen.ai pricing

Platform plans (data storage and repositories)

PlanMonthlyYearlyIncludedCost per GB stored (monthly)
ExplorerFreeFreeUnlimited public repos with unlimited collaborators; 5 private repos (max 3 collaborators); 50 GB storage; 50 GB transfer-
Hacker$30/month$300/yearUnlimited private repos; 100 GB storage and 100 GB transfer (more available)$30 / 100 = $0.30
Pro$60/month$600/yearEverything in Hacker; 500 GB storage and 500 GB transfer (more available)$60 / 500 = $0.12

The yearly toggle is labelled "Save 8%", but $300 a year against $30 a month is a saving of about 17%. Extra storage pricing is not listed. Non-profit and academic organizations can contact Oxen for a discount.

Models are pay as you go, with no subscription needed. Examples from the pricing page: Kling 4.0 text or image to video at $0.20 a second (720p) or $0.31 (1080p), so a 5-second 720p clip is 5 x $0.20 = $1.00; Seedance 2.5 reference to video from $0.12 a second (480p) to $1.34 (4K); FLUX 3 images from $0.0533 (768 square) to $0.7891 (4K); DeepSeek V4.1 Flash at $0.39 / $1.56 per million input / output tokens; Claude Fable 5.1 at $11 / $55.

Dedicated GPUs bill per second and shut down after 15 minutes of inactivity: A10G $1.65/hr, H100 (40 GB MIG) $1.95/hr, H100 $4.87/hr, H200 $9.98/hr, 4x H100 $15.00/hr, 2x H200 $19.50/hr, 8x H100 $38.99/hr. Fine-tuning is billed per second of training on the listed GPU (for example FLUX.3 Klein or Qwen Image 2.1 LoRA on 1x H100 at $4.87/hr), then the model can be deployed at the same dedicated-GPU rate.

Prices checked on the vendor's pricing page on November 2, 2026. See Oxen.ai pricing.

Key features

What Oxen lists on its site and docs:

  • Git-like branches, commits, diffs and merges for datasets, weights and media
  • Open-source CLI, Python package and self-hostable Oxen Server (Apache 2.0)
  • Tabular, text and line diffs, plus data frame tools in the Python API
  • One inference API for 200+ text, image, video and audio models
  • Prompts, references and generations stored with lineage, tags and folders
  • No-code and API fine-tuning (full or LoRA) on your own datasets
  • Dedicated GPU endpoints from A10G to 8x H100, billed per second
  • Team collaboration with automatic versioning and rollback

Who Oxen.ai is best for

Oxen fits ML engineers and small AI teams who need to version large, changing datasets and want generation, fine-tuning and data history in one place, and creative teams producing lots of AI images and video who need to keep track of which prompt and model made which asset. The free tier is enough to try the version control on public data; the model and GPU costs scale with use.

If you mainly want an open model hub and community, Hugging Face is larger. For pure model inference APIs, Replicate and fal are the common choices. If you only need data versioning on top of your own storage, compare DVC and lakeFS.

What users say

We found no established review-site ratings for Oxen.ai to summarise. The open-source core has about 1,200 stars on GitHub and was still receiving commits on November 2, 2026, which suggests an active but niche developer following compared with larger ML platforms. Oxen's speed comparison with git-lfs comes from Oxen's own benchmarks, not an independent test. Try the free Explorer tier with a real dataset before moving production data.

Pros

  • Open-source, self-hostable data version control for very large files
  • Free tier with 50 GB and unlimited public repositories
  • Pay-as-you-go access to 200+ models with per-second video pricing
  • Fine-tuning and dedicated GPUs with idle auto-shutdown

Cons

  • Extra storage beyond plan limits is not priced on the page
  • Yearly discount label (8%) does not match the listed prices
  • Free private repos capped at 5, with 3 collaborators
  • Little independent review coverage

Frequently asked questions

How much does Oxen.ai cost?

Explorer is free (50 GB storage, 5 private repos). Hacker is $30 a month or $300 a year with 100 GB, and Pro is $60 a month or $600 a year with 500 GB. Models, GPUs and fine-tuning are pay as you go on top. Prices checked on the vendor's pricing page on November 2, 2026.

Is Oxen open source?

The Oxen version control tool, CLI and server are open source under the Apache 2.0 license on GitHub, and you can self-host them. The hosted platform with models, fine-tuning and GPUs is a paid cloud service.

How is Oxen different from Git LFS or DVC?

Oxen is a separate version control system built for large data, rather than an extension to Git. Oxen says it is 20 times faster than git-lfs in its benchmarks, and it adds dataset diffs, a hosted UI and model tooling. DVC keeps data in your own storage alongside Git.

Can I fine-tune models on Oxen.ai?

Yes. Oxen supports full and LoRA fine-tuning of open models across text, vision-language, image and video, billed per second of GPU time, for example $4.87 an hour on one H100. The tuned model can then be deployed on a dedicated GPU.

Which AI models can I use through Oxen.ai?

Oxen lists 200+ models, including Kling 4.0, Seedance 2.5, Veo 3.1, FLUX 3, Nano Banana 2, GPT Image 2, Qwen Image, Wan 2.7, DeepSeek V4.1 Flash and Claude Fable 5.1, all billed per use.

Do Oxen GPUs keep billing when idle?

Dedicated GPUs are billed per second of use and shut down automatically after 15 minutes of inactivity, according to Oxen's pricing page.

Compare Alternatives

Oxen.ai

Free plan · from $30/mo · API

Enjoying this? Get one like it in your inbox each morning.

one email a day · unsubscribe in two clicks · no third-party tracking