Reka AI Review: Can It Compete With More Popular Models?

If you’re here to read my Reka AI review you’re probably on the path of starting your own AI project. The good news is that it won’t be that difficult.

Reka Playground allows you to quickly (and for free) test the model capabilities, while well-documented API lets you deploy the solution even quicker.

I’ve conducted a series of tests (inspired by Reka Showcase) to give you a neutral overview of what this LLM can and can’t do.

What is Reka AI?

Reka AI is a multimodal LLM (Large Language Model) focused on processing videos, images, audio, and text.

The focus of the company is to help developers and investors build solutions by using the power of multi-purpose models.

Features

Reka AI comes with three models:

• Core (largest and most capable)
• Flash (optimized for everyday use)
• Edge (smallest and fastest model)

I’ll use Core for all the tests, so you can get the idea of what you get with the most powerful model (if this one doesn’t satisfy you, Flash and Edge won’t also.)

Reka Playground

Like any other LLM, Reka AI comes with a minimalistic in-browser chat interface. There’s not much to think about UI/UX-wise, since the standards are already set, and Reka AI is successfully following them. If you’ve ever used ChatGPT, you’ll know what I’m talking about.

Previous conversations are stored in the left sidebar, with the option to delete the whole history at once. Individual chats can also be deleted, but also shared and their title edited.

As for the input options you can send:

• Textual prompts
• Images (up to 12MB)
• Videos (up to 30MB and 1 minute)
• PDF files
• Audio recordings (directly through microphone)

All files you send are stored in your account, so you can re-use them. The option to clear file history is available, but I find it strange that I couldn’t delete individual files.

Reka AI as an AI chatbot

The main thing you have to select is the model you’ll use (as described earlier.) I’ll focus on the most powerful one (Core), and see what results I can get.

Prompt: Generate a cooking landing page code ready for use.

As a response, I got code for three files (index.html, style.css, and script.js). Once I pasted them on my PC and ran them, this is what I saw:

Images are missing, but overall the result is decent. Now I also wanted to have a catchy poem on this topic.

Prompt: Write me a catchy poem on this same topic. Make it short and sweet, like cake!

Here’s the output:

In kitchens where magic stirs and whisks dance,
Beneath the warmth of ovens’ gentle trance,
Flour clouds kiss the air, spices sing their tune,
As culinary dreams in every pot commune.

From sizzling pans to bubbling pots of cheer,
Cooking delights, oh, how they draw us near!
With every slice, every bite, a story told,
In flavors bold, in moments sweetly bold.

So roll up your sleeves, let your creativity flow,
In the art of cooking, you’re the star, you know.
For in every dish, love and care are baked,
In every meal, memories gently awaked.

Standard AI-generated poetry – the best way to describe what I got. Anyways, since Reka AI is focused on image recognition features, let’s see in more detail how it does that.

Image Captioning

A while ago I tested some AI painting generators, so I decided to provide the best sample to Reka AI. As a result, I’m expecting a detailed explanation of what’s happening in the picture.

Tip: You can potentially use this to create image captions for your blog!

After uploading the output generated with Midjourney, I got a detailed response.

The image is a vibrant and dynamic painting that captures a tropical island scene at sunset. The canvas is filled with a rich palette of colors, with the sky painted in a blend of warm oranges and yellows, transitioning into cool blues and greens. The sun is depicted as a large, glowing orb, casting a warm light over the scene. The ocean is rendered in a mix of blues and greens, with the waves depicted in a lively, textured style.

The island itself is home to a single palm tree, its leaves swaying in the breeze. The painting is done in an abstract style, with visible brushstrokes adding to the overall sense of movement and energy. The perspective is from the ocean, looking towards the island, giving the viewer a sense of being in the scene. The painting evokes a sense of tranquility and beauty, capturing the essence of a tropical paradise.

Although it was very accurate, I wanted something shorter.

I asked for it, and in a moment I got the following result:

This is an abstract painting of a tropical island at sunset. The sky is a blend of warm and cool colors, with the sun casting a warm glow. The ocean is depicted in blues and greens with textured waves, and there is a single palm tree on the island. The perspective is from the ocean looking towards the island.

As you can see, it performed well!

Location Recognition

Ever played a geo-guessing game? Well, be careful you do next time since your opponents can use Reka AI to get an unfair advantage.

I tried first uploading a landmark of a 200,000-person city in Europe (Banja Luka), but it failed to recognize it.

Then I went with a landmark from another city with a population of close to 2 million inhabitants (Belgrade), and this time it could be identified that it was Eastern Europe. However, even though I was a bit pushy about the Reka AI model taking a guess, it stopped there.

After that, I took a random Google Street View screenshot, and the model guessed on the first try it was from Amsterdam.

As you can conclude, it’s not perfect, but it can get the job done if given enough details.

Food Recognition

For the food recognition test, I’ve decided to provide one image of the following dishes:

• McDonald’s burger
• Sarma
• Chicken fried rice

Each test was completed with the following prompts (in the same chat):

Prompt #1: Do you recognize this dish and can you tell me where does it come from?

Prompt #2: Describe shortly how healthy it is.

Prompt #3: Is it safe for a vegetarian to eat it?

Here are the results!

McDonald’s burger was identified as a cheeseburger from the US. As for health, it told me that it’s rich in proteins, vitamins, and minerals while containing a lot of saturated fat and cholesterol. It also noted that it’s not suitable for vegetarians.

For Sarma, it told me it was ‘stuffed cabbage rolls‘ from Eastern Europe, the Middle East, and Asia, which is a very close identification. Health-wise it provided a fair estimate, concluding with ‘if the filling contains high amounts of fat‘, which makes a lot of sense since it can’t see it inside, which also makes it impossible to answer if it’s suitable for vegetarians.

Finally, for the last dish the model identified it as ‘… a plate of fried rice, a popular dish in many Asian cuisines…‘ The health conclusion was that it depends on the type of rice it’s prepared with. The verdict on this dish being vegetarian-friendly was indecisive, since the model failed to notice the chicken on the plate.

Data Understanding

Reka AI can be a potentially good choice if you need a tool that can understand graphs and tables. I took one of the graphs generated with an AI for data visualization and asked a relevant question.

Prompt: This graph shows how many cellphones is lot by age. What age group is most likely to lose the phone?

Reka AI accurately replied that ‘The age group most likely to lose their phone, according to the graph, is the 25 to 34 age group…’

I sent a weather graph from the National Weather Service with the following prompt:

What is the warmest and what is the coldest month of this year?

Once again, the model successfully identified that ‘The warmest month of the year, according to the graph, is July…‘ and that ‘The coldest month appears to be January…‘

Notice how it avoids claiming anything with certainty since the data provided on charts can also be false. Brownie points for that!

Finally, I took a screenshot of a table of head-to-head comparisons between Adobe Firefly and Midjourney, and paired it with this prompt:

I'm just starting with AI-generated art and I don't have any budget to start with. What do you recommend me between these two solutions?

The answer was just absolutely correct!

If you’re just starting with AI-generated art and don’t have a budget, I would recommend starting with Adobe Firefly 2. It offers a free version with 25 generative credits, which allows you to create a few pieces of art without any cost. Additionally, it has a native GUI that can make the process of generating art more intuitive and user-friendly. While the photorealistic output is limited, it can still provide a good starting point for exploring AI-generated art.

Let’s see what else it can do!

Street Signs Comprehension

Reka AI can be a swift way to get yourself into traffic automatization! Or that’s what its creators want you to believe. I’m a bit skeptical, so I’ll provide several samples of street signs and see what it has to say about them.

There was a lot of hype about the cow UFO sign, so I decided to start with that.

Prompt: What is meaning of this street sign?

Here’s the answer

This street sign is a humorous take on the common yellow warning signs found in rural areas, which are meant to alert drivers to the presence of livestock on or near the road. The addition of the UFO above the cow adds a playful twist, suggesting that the cow may have been involved in an extraterrestrial encounter, which is a reference to the popular culture trope of UFOs and alien abductions, often associated with cows. It’s not an official traffic sign but rather a novelty sign meant to entertain passersby.

The next one comes directly from a driving school test.

The model replied correctly that the answer is B.

Finally, I wanted to confuse it a bit, so I’ve put an image showing that Highway 405 leads both to Sacramento and LA Airport.

The model told me that ‘…The sign in the image indicates that the I-405 leads to both the airport and Sacramento, which is accurate…‘ with a further explanation that ‘…the I-405 has an exit for the airport, which is why it is listed as a destination on the sign…’

Sure, Reka AI isn’t going to replace Tesla and its data-fed AI model anytime soon. However, it’s still a great and efficient way to quickly get into the (amateur) industry without too many complications.

Screenshot Analysis

I jumped right into screenshot analysis and was disappointed immediately. I sent a screenshot from my AC app where it’s indicated that the AC was outside the network. Then, I asked Reka AI why I couldn’t turn it on, and it immediately told me it couldn’t say what the cause was.

Still, I kept my optimism and I decided to check with a Fruit Ninja screenshot I have below:

Explanation: Fruits pop up on your screen and you have to ‘slice’ them by swiping your finger. In the Classical mode, you can miss up to three fruits. I did exactly that and the moment before the game over screen appeared I took this screenshot.

Prompt: Am I close to taking the high score? What are my chances?

The answer, this time wasn’t fully satisfying, but it was much better than the previous time.

…the current score of 16 and the best score of 319, you are significantly below the best score. To take the high score, you would need to score at least 204 more points. The chances of taking the high score would depend on several factors…

Reka AI failed to see that I’ve already used all of the available lives but to be fair, it’s also not indicated. I’ll give it a pass here.

Prompt: I want to get to Paris going through as few countries as possible. What is my best route from the ones you see on the screen?

The answer is more than disappointing.

The best route from the ones you see on the screen that goes through the fewest countries is the one that starts in Paris, goes through Belgium, and then to the Netherlands. This route goes through only two countries.

Things aren’t looking good for this Reka AI feature at this point!

Other Applications

It seems that the Reka AI team went a bit overboard with some showcases. But let’s see how the tool performs across some additional use cases.

Here’s how Reka AI explained the meme I provided to it.

This meme is a humorous take on the common experience of trying to teach older generations how to use technology. The character is confused about the concept of the “any” key, which is a reference to the outdated practice of pressing any key to proceed with a computer setup or installation. The meme plays on the idea that older individuals might not be familiar with the nuances of modern technology and the changes that have occurred since the time when pressing any key was a necessary step.

So far so good!

Another more practical use case could be discovering brand logos. I provided an image which contained half of the logo and asked

Tell me which company is the sponsor of this event.

It didn’t face any obstacles in detecting it!

As the final test, I wanted to find out if Reka AI is good at counting. I’ve provided the image below with three seated passengers and asked:

Based on this image, tell me how many passengers are on this bus.

Although one of the passengers is barely visible, due to the black clothing, Reka AI successfully identified all three of them.

Deployment Options

Besides the browser-based chat, Reka AI models can be used locally and through an API.

API keys are easily created from the Reka AI dashboard. The only note here is that you’ll have to have active credits (more about this in a bit.)

Out of the box you have Python SDK and HTTP API, which are fully documented on the Reka API documentation page.

All of the input options stay available even if you’re using API and there’s an option for API versioning. Currently, there are two versions available.

You get twice as many models since each one also has a twin which is more recently released (ex. reka-core and reka-core-20240415.)

If you want to have Reka AI locally on-premise, you’ll have to get in touch with their team as they didn’t disclose any details publically (besides the fact it’s possible.)

How to use Reka AI?

Here’s a quick tutorial to help you get started with Reka AI:

1. Head to Reka AI and hit the Try Our Models Free button.
   
2. Hit the Sign Up button and register using your email address.
   Alternatively, you can use your Google account.
   
3. Select the model (Core is the best), type your prompt, and hit the send button.
   
4. (Optional) Click on the Attach a file button and select a suitable format.
   
5. (Optional) Once the attached file is uploaded, type your prompt and hit the send button.

That’s it! Using Reka AI Playground is as easy as it gets, especially if you’ve already used similar tools like ChatGPT or Gemini. However, this interface is intended for testing purposes…

If you want to use the full potential you should go with API. You’ll have to navigate to the API keys section of your dashboard and from there you can create new keys. However, they’ll be usable only if you provide billing details and top-up your account balance. More about that in the section below.

Pricing

Reka AI offers a free chat (Reka Playground) environment in which you can test all three available models. Since the goal is to sell bulks of tokens for hosted API purposes, the company doesn’t charge for the Reka Playground usage*.

*Some limits do exist, however, if you intend to use it fairly, you’ll likely not reach them.

Reka AI models pricing:

• Core ($3 per 1M input tokens & $15 per 1M output tokens)
• Flash ($0.8 per 1M input tokens & $2 per 1M output tokens)
• Edge ($0.4 per 1M input tokens & $1 per 1M output tokens)

As the payment gateway, Reka AI uses Stripe, meaning you can any major credit or debit card to buy credit.

Upon toping up your Reka balance, you’ll be able to activate API keys and immediately use them in your projects.

Transactions you make to purchase the credits are non-refundable.

All of the tests conducted for this review were done using the Core model through Reka Playground. I only had to register for an account and didn’t have to pay or provide any billing details. This for me constitutes an amazing free trial.

Reka AI Review – Verdict

So after seeing all of this, I conclude that Reka AI can indeed compete with more popular models like ChatGPT and Gemini. However, since it’s purely oriented on larger scale use-cases, rather than in-browser chatting, it’s not going down the path of getting mainstream attention.

To conclude this Reka AI review I want to highlight its image understanding capabilities. Regardless of your goal and project, it can provide a lot of value for captioning, recognition, understanding, and comprehension of various situations.

Learn more about how the most powerful AI language models stand to one another in my Meta AI vs Gemini vs ChatGPT comparison.