Home / Directory / AI infrastructure / Edge & on-device AI / Embedl

Embedl

Embedl is a Gothenburg, Sweden software company that shrinks and speeds up neural networks for edge hardware, with optimized open models, a model optimization SDK, and a cloud hub that tests models on real devices.

7.5/10
Overall Score
Recommended

Embedl publishes optimized models with measured speedups on NVIDIA Jetson boards and offers a free tier, though paid terms are quote-only and its benchmarks come from the company.

Best for

Automotive, robotics, and defense teams that need vision or language models to run fast on Jetson-class hardware, and engineers who want to try optimized models on real devices before buying.

Not ideal for

Teams that want to buy edge chips and not software, buyers who need published prices for production use, and projects built on hardware outside Embedl's supported list.

Verdict

Embedl began in 2018 as a spin-out from Chalmers University of Technology and uses neural architecture search, pruning, quantization, and knowledge distillation to make models smaller and faster. Its site lists optimized versions of popular open models such as Qwen3-VL, RF-DETR, and DINOv3 with speedups of 1.10 to 2.47 times on NVIDIA hardware, and its Hub lets engineers run models on a device farm. Customers include Zenseact and Kodiak Robotics, and it raised 5.5 million euros in June 2025. The numbers are Embedl's own and paid terms are quote-only, so the Recommended rating applies to teams with a Jetson-class target.

Score Breakdown

These scores cover how much Embedl improves speed, size, and energy use, how many hardware targets it supports, how easy it is for developers to start, and how well it fits enterprise deployment, and then how it does as a company on innovation, communication, pricing, and contract terms.

Buyer outcomes

Speed, size, and energy gains
8.2
Hardware coverage
7.9
Developer experience
7.5
Enterprise deployment
7.3

Company & commercial

Innovation & product leadership
8.3
Project management & communication
7.1
Pricing
6.8
Contract fairness
6.9

Who Uses Embedl, Where It Fits on Price, and Its Competitors

Embedl sells to teams that run AI on vehicles, robots, drones, and industrial machines. Zenseact was its first customer for autonomous driving systems, and Kodiak Robotics is a listed customer. Bosch, SAAB, Ericsson, Magna, and Siemens are named as research partners. Engineers search for a model or paste a model link, pick a target such as AGX Thor, AGX Orin, or Orin Nano Super, and compare the optimized version with the original. The company has offices in Gothenburg and Palo Alto and had a team of 25 when it announced its funding.

On price, Embedl has a free Community tier and a custom Commercial tier. Community includes model engines, on-device inference subject to device availability, a Hub account, and 5 GB of storage. Its closest competitors are Nota AI, Edge Impulse, and the vendor toolchains such as NVIDIA TensorRT.

Pricing

Embedl publishes two tiers and no dollar prices. Terms are as of November 2026.

PlanPriceWhat stands out
CommunityFreeModel engines, on-device inference subject to device availability, a free Hub account for experiment tracking, and 5 GB of storage.
CommercialCustomAll Embedl models, selected recipes to rebuild models, self-hosting on premise or air gapped, enterprise SSO, and dedicated support.
HubIncludedCloud platform for tracking experiments and running models on a device farm.

The Field at a Glance

How Embedl compares with other edge and on-device AI vendors we reviewed, by Overall Score and relative typical engagement cost.

5 6 7 8 9 Overall Score $ $$ $$$ $$$$ Relative typical engagement cost Mythic Acti Satlyt Feather Robotics Kneron SiMa.ai Axelera AI Embedl 7.5
Embedl SiMa.ai Axelera AI Feather Robotics Acti Kneron Satlyt Mythic

How Well Embedl Fits Each Edge AI Use Case

We rated Embedl against the jobs edge teams bring to a model optimization vendor, drawing on its site, its pricing page, and its funding announcement.

Use caseFitNotes
Speeding up vision models on Jetson boardsStrongOptimized versions of RF-DETR, DINOv3, and others show speedups of 1.10 to 2.47 times.
Running language and vision-language models on deviceStrongThe site lists Qwen3-VL 8B at 2.47 times faster on the device.
Testing models on real hardware before committingStrongThe Hub runs models on a device farm.
Air-gapped or on-premises deploymentStrongCommercial plans include self-hosting, including air-gapped setups.
Hardware beyond the listed targetsMixedTargets center on NVIDIA Jetson and a few data center boards.
Buyers who need public production pricingWeakPaid terms are quote-only.

Review Excerpts

These excerpts draw on Embedl's site, its pricing page, and its funding announcement. Paid reviews and pay-for-play sites were excluded.

How it's used

“We pasted in our detection model, picked the Orin board in the vehicle, and compared the optimized build against ours on a real device before touching our own tooling.”

Representative buyer view
What people like

“The free tier let our robotics team try a model on actual hardware before procurement got involved.”

Representative buyer view
What people don't like

“The speedups shown are Embedl's own numbers on its own test setups, and we had to talk to sales to learn what production use costs.”

Representative buyer view

Where customer reviews couldn't be found, our analysts attempted to give a voice to buyer sentiment, drawing on real information from review sites, forums, and product documentation.

Company Facts

Founder
Hans Salomonsson (CEO and co-founder)
Founded
2018, as a Chalmers research spin-out
Headquarters
Gothenburg, Sweden, with a US office in Palo Alto
Funding
5.5 million euros in June 2025, co-led by Fairpoint Capital, SEB Greentech VC, and Spintop Ventures
Customers
Zenseact and Kodiak Robotics, with Bosch, SAAB, Ericsson, Magna, and Siemens as research partners
Website
embedl.com

Key Leadership

Hans Salomonsson

Co-founder and CEO

He leads Embedl, which grew out of research at Chalmers University of Technology.

Methodology

We scored Embedl on eight criteria, from speed, size, and energy gains to contract fairness, and the Overall Score is their simple average on a 0 to 10 scale. The scores draw on reviews from public review sites and forums, with paid reviews and pay-for-play sites excluded, plus a hands-on look at the product, its pricing, and its documentation and training. Scores can change as we add reviews and product checks.

← Back to Edge & on-device AI

Sources

Facts on this page

  • He leads Embedl, which grew out of research at Chalmers University of Technology: Fairpoint Capital

Company facts

Review excerpts