Kimi k1.5 vs DeepSeek R1: Battle of the Best Chinese LLMs

It’s the era of Chinese supremacy in generative AI, and we love it! Yet another notable Chinese company, Moonshot AI, has just released its latest version of the Kimi k series models – Kimi k1.5. This open-source, multimodal LLM is a strong competitor to the popular models by Open AI, Claude, Qwen, and Deepseek. With advanced image understanding, text generation, and reasoning capabilities, Kimi k1.5 is surely making headlines across the generative AI space. It is free to use and available on their chat interface. In this blog, we will test its capabilities against DeepSeek-R1 – a model that has been topping the charts across various benchmarks. Let the Kimi k1.5 vs DeepSeek-R1 battle begin!

What’s Kimi k1.5?

Kimi k1.5 is the latest LLM by Moonshot AI, a Chinese AI firm founded in 2023. It is an open source, multimodal model with an enhanced 128 K context window that enables it to process large amounts of information in a single prompt. The model is completely free to use with no limits! Kimi k1.5 shows great potential at tasks involving STEM, coding, and general reasoning. It outshines giants like OpenAI o1, OpenAI o1-mini and Qwen models like QVQ-72B/32B Preview on several parameters like Maths, Coding and Vision.

Key Features of Kimi k1.5

1. Unlimited Use for Free: The model is completely free to use and with no usage limits.
2. Web Search at Scale: It can perform real-time web search across 100+ websites.
3. Multiple Files at Once: It can analyse up to 50 files including PDFs, docs, PPTs and even images in a single go with complete ease.
4. Advanced Reasoning: It showcases advanced chain of thought reasoning capabilities.
5. Enhanced Image Analysis: Its image analysis skills go beyond basic text extraction. It can actually answer questions by understanding the context of images.
6. Set Common phrase: It allows you to set up common phrases, so that you don’t same to write the same prompt multiple times.

How to Access Kimi k1.5?

To access the Kimi k1.5 model, follow the below steps:

1. Head to https://kimi.ai/.
2. To access this model, you will have to create your account. In the centre of the screen, at the left side, click on “log in”.
3. On the home page, below the chatbox, on the left hand side, click on “Kimi”. From the dropdown list, select “K1.5 Loong Thinking”.

What is DeepSeek-R1?

DeepSeek-R1 is the latest LLM by Chinese AI startup, DeepSeek, which too was founded in 2023. Since its launch a week ago, this model has shaken the GenAI world with its capabilities, giving paid models of OpenAI and Claude a run for their money. It is also an open source model that showcases amazing reasoning, coding, and mathematical skills.

How to Access DeepSeek-R1?

To access DeepSeek-R1 follow the below steps:

1. Go to https://chat.deepseek.com/.
2. Sign up to create your account.
3. In the middle of the screen, click on “DeepThink”.

Also Read: DeepSeek R1 vs OpenAI o1 vs Sonnet 3.5: Battle of the Best LLMs

Kimi k1.5 Vs DeepSeek-R1

Now let’s explore the capabilities of both these models. I will give the same prompt to both of them and compare the outputs, evaluating them on various skills like  image analysis, web search, handling multiple files, coding and logical reasoning. Lets start.

Task 1: Image Analysis

Prompt:  “Go through the two images and solely based on the images give me an analysis of how DeepSeek-R1 performs against Kimi k1.5 long-CoT”

Image1 Image 2

Note: While using Kimi k, at the center of the screen, under the chatbox, click on “online” to shift the model to offline mode. This ensures that it doesn’t take any help from the internet, and gives an analysis solely based on the images.

Output:

DeepSeek-R1

Kimi k1.5

Review:

I expected the LLMs to just compare the common parameters shown in the two images for DeepSeek-R1 and Kimi k1.5. But both the models compared the parameters for which information was not provided. Yet, if we look at the numbers from solely a mathematical standpoint, both the models handled the numbers correctly.

Result:

Ideally, both the models have failed at this test. But Kimi k1.5 showcased better analysis of the text in the images compared to DeepSeek R1.

Score: Kimi k1.5: 1 | DeepSeek-R1: 0

Task 2: Web Search

Prompt: “Find me the links for a red gown, under $200”

Note: While using Kimi k, at the center of the screen, under the chatbox, click on “offline” to shift the model back to online mode, ensuring it uses the web. In DeepSeek, remember to select the “search” option in the chatbox, to allow the model to access the web.

Output:

DeepSeek-R1

Kimi k1.5

Review:

I just wanted a list of websites that I can quickly access to find the red coloured gown within my budget. DeepSeek gave me a lot of options in the result, although none of them were directly relevant to me. Kimi k1.5 gave me limited options in the direct result and several options in the side panel. Although the two chosen links were the most relevant and useful, the additional panel listings gave me access to other websites I could refer to!

Result:

Kimi k1.5 stands out in this task for giving crisp and relevant results.

Score: Kimi k1.5: 2 | DeepSeek-R1: 0

Task 3: Handling Multiple Files

Prompt: “Summarise the contents of each file in brief”

Attachemt: Files

Output:

DeepSeek-R1

Kimi k1.5

Review:

DeepSeek could not process all the files at once and even after multiple attempts gave the same result. But when it was given each of these files, one by one, in different prompts, it gave good results. Kimi k worked seamlessly with all the input files. Although it gave a detailed summary of the PPT and the PDF, it didn’t account for the image in its result.

Result:

Kimi k1.5 processed 2 out of the 3 files and gave a comprehensive result.

Score: Kimi k1.5: 3 | DeepSeek-R1: 0

Task 4: Coding

Prompt: “Write the HTML code for a simple snakes and ladders game for 2 players”

Output:

DeepSeek-R1

Kimi k 1.5

Review:

The game interface generated by both the models were quite similar. In DeepSeek-R1’s output I could actually see the players moving across the board. In case of Kimi k1.5’s output, the players were moving outside of the board which didn’t really give the actually feel of the game. Overall, both the outputs lacked the core elements of “snakes and ladders” which are “snakes” and “ladders”.

Result:

DeepSeek R1’s code was more advanced and offers more flexibility. Its final interface was more fun to play with too.

Score: Kimi k1.5: 3 | DeepSeek-R1: 1

Final Score

Kimi k1.5: 3 | DeepSeek-R1: 1

Comparison Between DeepSeek-R1 vs Kimi k1.5

1. Performance and Speed

• DeepSeek-R1: Optimized for high-speed processing, delivering faster response times for complex queries.
• Kimi K1.5: Focuses on balanced performance, ensuring consistent speed across various tasks.
• DeepSeek-R1: Excels in handling large datasets with minimal latency.
• Kimi K1.5: Prioritizes stability over raw speed, reducing the risk of performance drops.

2. Accuracy and Precision

• DeepSeek-R1: Uses advanced algorithms to achieve higher accuracy in data analysis and predictions.
• Kimi K1.5: Relies on robust error-checking mechanisms to maintain precision in outputs.
• DeepSeek-R1: Better suited for tasks requiring fine-tuned results, such as scientific research.
• Kimi K1.5: Performs well in general-purpose applications with reliable accuracy.

3. User Interface and Experience

• DeepSeek-R1: Features a sleek, modern interface designed for tech-savvy users.
• Kimi K1.5: Offers a more intuitive and user-friendly interface, ideal for beginners.
• DeepSeek-R1: Provides customizable dashboards for advan users.
• Kimi K1.5: Focuses on simplicity, ensuring ease of use for non-technical users.

4. Scalability and Flexibility

• DeepSeek-R1: Highly scalable, capable of handling growing data demands effortlessly.
• Kimi K1.5: Offers moderate scalability, suitable for small to medium-sized enterprises.
• DeepSeek-R1: Supports integration with a wide range of third-party tools and platforms.
• Kimi K1.5: Limited flexibility but ensures seamless operation within its designed scope.

Conclusion

Kimi k1.5 is an exciting new model that showcases a lot of potential to be the next big thing in the world of conversational AI. It’s quick, efficient and can take in a large amount of context. Moreover it provides a well researched answer accessing different links across the web. DeepSeek-R1 on the other hand, captures attention with its detailed responses but falters when it comes to web search and handling larger chunks of data.

However, the LLM race, started by US-based companies, is now getting heated up, as their Chinese counterparts are releasing one stand-out model after the other. As these companies battle to the top, it’s just great that users, developers and companies get access to the latest and the most advanced technologies!

Ready to unlock the full potential of DeepSeek? Join our course now and master AI-driven analysis and automation to elevate your skills!

Also Read:

• DeepSeek R1 vs OpenAI o1: Which One is Faster, Cheaper and Smarter?
• DeepSeek R1 vs OpenAI o1 vs Sonnet 3.5: Battle of the Best LLMs

Frequently Asked Questions

Anu Madan

Anu Madan has 5+ years of experience in content creation and management. Having worked as a content creator, reviewer, and manager, she has created several courses and blogs. Currently, she working on creating and strategizing the content curation and design around Generative AI and other upcoming technology.