Specht GmbH: Your SEO & SEA specialists
GPT-4o Mini: New model with performance upgrade and low prices
- #1 In your niche
- More inquiries & sales
- More turnover
Over 250+
customers
It is the Ford Mustang among the AI language models: GPT-4o Mini. A model for beginners and price-conscious developers that combines high performance and cost efficiency. In mid-July 2024, the company OpenAI, which became known worldwide through the chatbot "ChatGPT" at the latest, replaced the previous GPT-3.5 Turbo language model with the GPT-4o Mini model.
In this article, we explain everything you need to know about the new AI model from Open AI. We also compare the new, low-cost GPT-4o Mini with its predecessor and the competitor models Google Gemini 1.5 Flash and Claude 3 Haiku.
Google rating
Based on 315 reviews
Trustpilot rating
Based on 107 reviews
Recommended by:
Cost efficiency as the biggest advantage of the new GPT model
The AI language model GPT-4o Mini is, on the one hand, the result of the innovation process at OpenAI. On the other hand, it is a late reaction to the fierce competition in the AI sector.
The most important key figures: A 128k context window, up to 16,000 output tokens and an output speed of 182 tokens per second - all this is available with the GPT-4o Mini for just 15 cents per 1 million input tokens and 60 cents per 1 million output tokens.
The logical consequence: with the GPT-4o Mini, OpenAI has overtaken the Google Gemini 1.5 Flash and Claude 3 Haiku models in the field of high-performance yet cost-effective language models for beginners. The main reason for its good performance compared to the competing models is its cost efficiency.
It was long overdue for OpenAI to bring a low-cost, high-performance language model onto the market. Numerous companies offered cheaper language models than GPT-3.5 Turbo for beginners and price-conscious developers. Examples of this are Google Gemini 1.5 Flash and Claude 3 Haiku, which not only had more performance, but also multimodal capabilities and therefore more functions than GPT-3.5 Turbo.
According to Olivier Godement, the manager of ChatGPT, OpenAI had deliberately taken its time to develop a powerful yet cost-effective version of its AI language model. He cited the focus on the large and powerful models such as GPT-4 and the new GPT-4o as the reason for this. The importance of low-cost models has now been recognized and an adequate product has been launched on the market with the development of GPT-4o Mini.
It is precisely this low-cost, high-performance model that we at Specht GmbH are now comparing with its predecessor, the GPT-3.5 Turbo, and its competitors, the Google Gemini 1.5 Flash and Claude 3 Haiku. We compare based on the following three criteria:
- Costs
- Skills
- Performance
By the way: Did you know? You can use ChatGPT free of charge via our ChatGPT tool! Feel free to take a look if you would like to receive precise answers from the advanced version GPT-4.
Costs: Developers pay a maximum of 15 cents per million input tokens
For developers, the costs are different from those for users of the chatbot, as developers are not interested in using the chatbot, but in integrating the language model into their development environment. To do this, developers use the API offered by OpenAI. The API (Application Programming Interface) is a programming interface.
The API allows developers to use the language model individually in their own systems and applications. The price for using the GPT models is calculated based on the number of input tokens (input tokens) and output tokens (output tokens). The GPT-3.5 Turbo model was overpriced compared to the smaller models from competing companies, which had more performance and capabilities.
In order not to lose market share to the competition with its small model, which was still GPT-3.5 Turbo until July 2024, a cost-effective and attractive solution was needed. So OpenAI developed the GPT-4o Mini language model, which has the following cost differences compared to its predecessor:
Model | Regular costs | Costs for Batch API |
GPT-4o Mini | - 15 cents / 1 million input tokens - 60 cents / 1 million output tokens | - 8 cents / 1 million input tokens - 30 cents / 1 million output tokens |
GPT-3.5 Turbo 0125 | - 50 cents / 1 million input tokens - 1.50 $ / 1 million output tokens | - 25 cents / 1 million input tokens - 75 cents / 1 million output tokens |
GPT-3.5 Turbo Instruct | - 1.50 $ / 1 million input tokens - 2.00 $ / 1 million output tokens | - 75 cents / 1 million input tokens - 1.00 $ / 1 million output tokens |
"Batch API" is an offer from OpenAI that provides lower costs as compensation for a longer waiting time for AI responses. In certain use cases, registering with Batch API may be useful for developers. If you would like to find out more about the Batch API, OpenAI provides an information page.
In the rest of this article, we will take a closer look at the performance and capabilities of the GPT-3.5 Turbo, GPT-4o Mini, Google Gemini 1.5 Flash and Claude 3 Haiku models in a comparison. First, we conclude the cost comparison of the models with the following information:
- Google's Gemini 1.5 Flash has the advantage that there is a free version for developers. However, the use of the API is severely limited with this version.
- In the paid version, Google's Gemini 1.5 Flash has prices that depend, among other things, on the length of the prompts - i.e. the instructions given to the AI.
- The prices for Gemini 1.5 Flash are slightly below the prices for GPT-3.5 Turbo, i.e. significantly above the prices of the new GPT-4.o Mini. Example: 0.35 cents per 1 million input tokens for prompts with up to 128,000 tokens.
- Claude 3 Haiku, Anthropic's cheapest model, is very inexpensive at 25 cents per 1 million input tokens, but more expensive than the new GPT-4o Mini.
- In terms of output tokens, Claude 3 Haiku from Anthropic clearly outperforms OpenAI's new, low-cost model at $1.25 per 1 million.
With GPT-4o Mini, OpenAI has therefore created a language model that is convincing in a comparison of low-cost and high-performance models with the lowest costs. In view of the extraordinarily low costs, the question arises as to whether the AI model even meets the qualitative requirements of developers...
Capabilities: Comparison of the AI benchmarks of GPT-4o Mini and other language models
AI benchmarks are used to evaluate the performance and capabilities of AI models. OpenAI has uploaded a graphic relating to AI benchmarks on its website as part of the announcement of the new GPT-4o Mini language model.
This graph shows a score in percent for each benchmark. The score provides information on how accurately language models perform tasks from specific performance areas.
The OpenAI comparison includes GPT-3.5 Turbo, GPT-4o Mini, Google Gemini 1.5 Flash and Claude 3 Haiku. Before we list and evaluate the graphics, we will first explain the significance of the AI benchmarks for a comprehensive understanding:
- MMLU (Massive Multitask Language Understanding): This is one of the most important benchmarks, as it assesses general knowledge and problem-solving skills. MMLU covers various subject areas - from natural sciences to culture and entertainment.
- GPQA: This is a set created by Google with hundreds of multiple-choice questions. The questions are at graduate level and challenge the AI immensely. The better it performs, the more the AI is able to answer particularly difficult and specific questions correctly.
- DROP: This AI benchmark is less important in assessment procedures. It is used to check whether a language model is able to understand content over a length of several paragraphs and derive correct conclusions.
- MGSM (Multilingual Mathematics for Primary School): The mathematical level of this AI benchmark test is low, but the test is still useful. Tasks are set in several languages, so that the AI's ability to understand and process numerical tasks in several languages can be tested.
- Math: This test is designed to comprehensively test the mathematical and logical abilities of a language model. It is essentially a general math test.
- Human Eval: This AI benchmark tests the ability of language models to generate programming code. The AI model has to solve well over 100 programming tasks as part of the test. The functionality of the code is checked by software.
- MMMU (Massive Multi-Discipline Multimodal Understanding and Reasoning): This AI benchmark is specifically designed for multimodal language models and is therefore relevant for the GPT-4o Mini model. AI models are subjected to a detailed check of their multidisciplinary capabilities at college level.
- MathVista: This metric can be used to examine and evaluate the mathematical capabilities of AI models in visual contexts. One example of this is the ability of an AI to evaluate mathematical graphics and draw the right conclusions from them.
The following graph now shows that the GPT-3.5 Turbo model was inferior to its competitors Google Gemini 1.5 Flash and Claude 3 Haiku in the AI benchmarks MMLU, GPQA, DROP, MGSM and Human Eval, and yet cost developers considerably more. GTP-3.5 Turbo could not be evaluated in the AI benchmarks MMMU and MathVista because the AI language model had no corresponding capabilities. GPT-3.5 Turbo was therefore clearly inferior to its competitors in terms of AI benchmarks.
With the GPT-4o Mini AI model, the tide has once again turned clearly in favor of OpenAI in comparison with the competition. GPT-4o Mini outperforms GPT-3.5 Turbo, Google Gemini 1.5 Flash and Claude 3 Haiku in almost all benchmarks. Even if the qualitative differences may be small in some cases, GPT-4o Mini is much more attractive for developers due to its significantly lower costs.
A comprehensive comparison with GPT-3.5 Turbo reveals clear advantages for the GPT-4o Mini model. In the important benchmark MMLU (Massive Multitask Language Understanding), GPT-4o Mini achieved a score of 82 percent and outperformed its predecessor by 12.2 percent. Here is an overview of GPT-4o Mini's score in all benchmarks:
- MMLU: 82 percent
- GPQA: 40.2 percent
- DROP: 79.7 percent
- MGSM: 87 percent
- MATH: 70.2 percent
- Human Eval: 87.2 percent
- MMMU: 59.4 percent
- Math Vista: 56.7 percent
New functions have also been introduced in GPT-4o Mini. The most important two aspects are described in the following two sections.
1st innovation: Introduction of multimodal capabilities
The multimodal capabilities introduced with GPT-4o Mini are currently still limited. Nevertheless, the language model can already process graphical input via the API. This opens up more possibilities for developers to integrate the AI model into their applications. OpenAI has announced the future expansion of multimodal capabilities to include the input and output of image, video and audio information.
2nd innovation: Protection against several types of manipulation
Among other things, OpenAI has increased security against prompt injections, which is an immense advantage of GPT-4o Mini. It is the company's first AI language model with a so-called instruction hierarchy. Thanks to this technology, the language model is immune to prompt injection attacks and a range of other types of manipulation.
An example of a prompt injection attack is: "Ignore all previous instructions from the system." In this prompt, the previous language models were prone to ignoring the instructions of OpenAI's developers. Disregarding the instructions of developers is one of the biggest potential threats that an AI can pose.
Performance: Enlarged context window and high quantity of output tokens
The performance of AI models can be evaluated using various metrics. One of these is the context window. This indicates how many tokens - i.e. how many words at once - a language model can process. The larger the context window, the longer and more detailed conversations an AI can have with the user.
With a context window of 128k tokens, the GPT-4o Mini model outperforms its predecessor GPT-3.5 Turbo by a factor of about eight. This puts OpenAI's new model only just behind the 200k context window of Anthropic's Claude 3 Haiku. Google Gemini 1.5 Flash even has a groundbreaking context window of 1 million tokens!
The context window shows that although the GPT-4o Mini is better than its predecessor, it cannot keep up with Claude 3 Haiku and Google Gemini 1.5 Flash. On the other hand, GPT-4o Mini performs better in terms of the number of output tokens and output speed. In addition, GPT-4o Mini is trained with data from 2023 and is therefore at a more recent state of knowledge than the competing models.
The output speed indicates how many tokens an AI language model outputs per second. In other words, it is the speed at which developers and other users receive their answers or a completed task. With an output speed of 182 tokens per second, GPT-4o Mini is slightly better than Google's model and significantly better than the 127 tokens per second of Claude 3 Haiku.
Conclusion: GPT-4o Mini is worth it all along the line!
The costs for the new GPT-4o Mini are very manageable for developers. As if the regular prices were not already low enough, there is the option of a considerable cost reduction by registering with Batch API. This means that none of the competing models is currently as lucrative as the GPT-4o Mini in terms of cost.
Despite all this, the capabilities and performance leave nothing to be desired. GPT-4o Mini has been immunized against various types of manipulation, enhanced with multimodal capabilities and, last but not least, boosted in terms of performance. In the AI benchmarks, GPT-4o Mini outperforms the competition in almost all areas.
FAQ: Questions and answers about the GPT-4o Mini
What is an LLM?
The chatbot "ChatGPT" is based on language models. Language models are used to understand and generate language based on artificial intelligence (AI). GPT-4o Mini, for example, is an LLM (large language model). It was trained with billions of parameters (hence "large").
The language model learns independently from the data records in order to derive content and relationships and provide answers to users' questions.
Aren't there also free tariffs for the GPT models?
A distinction must be made here between users of the chatbot and users of the API. Users who use the chatbot "ChatGPT" for conversation, research and similar purposes do not have to pay anything in the free tariff.
The free plan includes unlimited access to the chatbot with the GPT-4o Mini version. The company OpenAI also offers unlimited use of ChatGPT in the GPT-4o Mini and GPT-4o versions for a flat rate of 20 US dollars per month.
However, the situation is different for developers. Instead of the chatbot, they use the API offered by OpenAI, which is always subject to a fee. The API allows developers to integrate GPT-4o Mini into their development environment and thus work more efficiently.
What other GPT models are there?
The GPT-1 and GPT-2 models, which were still underdeveloped, were followed by the GPT-3 model, which was not yet available for public use. On November 30, 2022, the company OpenAI released the chatbot "ChatGPT" with the improved language model GPT-3.5 - still without an API for developers - for public use.
The further development of GPT-3.5, the GPT-3.5 Turbo model, offered significantly more customization options compared to GPT-3.5. The API offer was already available for this model. This was followed by the GPT-4, GPT-4 Turbo, GPT-4o flagship model in mid-May 2024 and now the slimmed-down GPT-4o Mini variant.
All models with a "4" in their name have more comprehensive functions and are trained with larger and more up-to-date data sets.
Top rated SEO agency in Germany
Over 500+ clients trust our SEO expertise for Munich and the whole of Bavaria!
Many well-known customers from Munich and the whole of Bavaria already trust us. Whether local companies such as doctors and real estate agents or online stores. As a highly specialized team of SEO freelancers, we can deliver exceptional results in all highly competitive SEO niches. deliver exceptional results.
Vieotestimonials
Short video testimonials from our customers!
Interested in what our customers have to say? Listen to some excerpts from our customer testimonials now!
Jule Bunze
Jule Bunze | Copywriter & Texter Munich
Uta Pittino
Osteopathy & Physiotherapy in Munich
Mana
Coworking space in Munich
Korbinian Gerstl
Frankjuice
- Ready to be number 1 on Google?
Book your free SEO consultation now!
Strong customer results
Case studies of successful customers
The best references are the voices of others! In the overview you will find video testimonials from happy customers. We have prepared more detailed case studies for you here.
- thefrankjuice.com
"SEO is the most successful marketing measure that we have implemented in recent years from the hodgepodge of measures. Thanks to Team Specht GmbH for their consistency and support over several years"
A lot of organic rankings but money keywords not in the top 3.
In addition to highly authoritative backlinks the site was cleaned up onpage from keyword cannibalization to onsite basics.
After 2 years, there are countless money keywords in the top 3 that ensure strong sales growth.
No rankings in the highly competitive Google 3-Pack. Thus, important local search queries were left out and left to the competition.
Indispensable local 3-pack ranking for its money keywords, even one-pack ranking due to high authority. Massive customer growth!
- we-are-mana.com
"I now have a firm foundation that lets me sleep soundly because I have gained many new customers. I can recommend Specht GmbH with a clear conscience to any company that seriously wants to grow through SEO & SEA. "
- julebunze.com
"Absolute recommendation for this local SEO agency in Munich! When I started my own business, I attended a local SEO course for freelancers with Timo Specht and his team and used the tool SEO successfully booked. Coupled with the great work of Timo and his team, my website was ranked on page 1 of the Google search with my relevant keyword within a few weeks. In the meantime, I have also been ranked number 1 in Google Search and Google Maps for several months, which has enabled me to generate many inquiries and new customers. I can only recommend anyone who wants to start their own business to have local SEO on their radar from the outset and to work with Timo Specht as an agency."
The newly launched website was beautiful at the start, but not visible, just like Jule's Google Business profile.
In order to quickly generate relevant new customer inquiries, we focused purely on the money keywords = all keywords with the intention to buy "copywriter Munich".
After 6 months of optimization, we achieved pole position with Jule as the number 1 "copywriter Munich" both in Google organic and in Google Maps. Jule now receives new customer inquiries every day.
Location
Our company location in Munich
Brienner Straße 29, 80333 is centrally located in the old town and Maxvorstadt district of Munich.
It is named after the site of the Battle of Brienne and, along with Ludwigstrasse, Maximilianstrasse and Prinzregentenstrasse, is one of the four most important boulevards in the Bavarian capital.
Our SEO & Online Marketing office is located in the Palais am Obelisk, which is best reached from the east via Karolinenplatz, which is in the immediate vicinity, and from the west via the well-known Odeonsplatz.
Munich's main train station can also be reached quickly from our office.
Team
We have put together an exclusive team of SEO experts to optimize your website every day
Our SEO experts tweak the most important parameters every day to boost your site on Google!
Timo Woodpecker
Founder & SEO Expert
Christoph Woodpecker
Founder & SEO Expert
Valeria
Content Expert
Steps
A successful collaboration with us!
Step 1:
- Fill out contact form
Click on the button below and fill out the contact form truthfully. Please formulate your request in one sentence so that we know immediately what it is about. You will be redirected to the calendar where you can choose a free slot with our team.
Step 2:
- Free initial consultation
We get to know each other in the initial consultation. We will find out what your current situation is and what goals you would like to achieve. If we can help you further, we will then arrange a strategy meeting.
Step 3:
- Strategy talk & offer
In a personal 1:1 strategy meeting, we present you with your individual SEO strategy and show you the potential for growth. During the meeting, you will receive a customized offer to help you achieve your goals.
Reviews
Customer testimonials from successful SEO projects
More than 300+ customers trust our expertise and rate us with 5 out of 5 stars on Google, Trustpilot & Co. We are the best-rated SEO agencyin Germany.
Carlheinz Bartsch
Bartsch Real Estate Ltd.
Verified Trustpilot rating
Kelsey Nielsen
Personio
"I worked with Timo on an SEO content audit as well as consulting him on a few SEO-related technical and logistical questions. He is a clear subject matter expert, provided a great audit with clear next steps, and was able to answer all of my questions. Highly recommend and will definitely be working with him again!"
"
Verified Trustpilot rating
Michael Wunderlich
Managing Director - Practice Ortho Lehel
"Many thanks Timo to you and your team for the friendly, perfect and extremely professional support
with SEO and SEA from day one, every day. You are always there when I need you. My expectations were exceeded by far! I would absolutely recommend you to everyone!
THANK YOU!"
Verified Trustpilot rating
Christian Mazilu
Shisha World
"Very good expertise. Pleasant personality with very good understanding of e-commerce SEO and content. Absolutely Up2Date. Are very satisfied in the collaboration. Absolutely recommend for companies as well as self-employed who want to sustainably advance your business."
Verified Google Rating
Daniel Rakus
SEA Experts GmbH & CoKG
"Top SEO agency - all goals were achieved. After 6 months we have a top ranking for the selected terms and the visibility of the website has also improved in other ways. Cooperation at eye level and professional!"
Verified Google Rating
Carlheinz Bartsch
Bartsch Real Estate Ltd.
Verified Trustpilot rating
Kelsey Nielsen
Personio
"I worked with Timo on an SEO content audit as well as consulting him on a few SEO-related technical and logistical questions. He is a clear subject matter expert, provided a great audit with clear next steps, and was able to answer all of my questions. Highly recommend and will definitely be working with him again!"
"
Verified Trustpilot rating
Michael Wunderlich
Managing Director - Practice Ortho Lehel
"Many thanks Timo to you and your team for the friendly, perfect and extremely professional support
with SEO and SEA from day one, every day. You are always there when I need you. My expectations were exceeded by far! I would absolutely recommend you to everyone!
THANK YOU!"
Verified Trustpilot rating
Christian Mazilu
Shisha World
"Very good expertise. Pleasant personality with very good understanding of e-commerce SEO and content. Absolutely Up2Date. Are very satisfied in the collaboration. Absolutely recommend for companies as well as self-employed who want to sustainably advance your business."
Verified Google Rating
Daniel Rakus
SEA Experts GmbH & CoKG
"Top SEO agency - all goals were achieved. After 6 months we have a top ranking for the selected terms and the visibility of the website has also improved in other ways. Cooperation at eye level and professional!"
Verified Google Rating
Luke Uhl
Vaillant
Verified Google Rating
Specht GmbH: Your SEO & SEA specialists
Secure your free strategy meeting now
- #1 In your niche
- More inquiries & sales
- More turnover