Increasingly tough price war among AI models
Because it is rather questionable whether many millions of end consumers will pay monthly subscription fees for chatbots on a permanent basis, providers of AI models have increasingly focused on optimizing their APIs for these AI models in the B2B business. For developers who integrate these APIs into their apps and web services, it is not only the quality of the output that is crucial, but increasingly the prices.
OpenAI recently introduced the GPT-4o mini, the little brother of its latest and best AI model. Not much is known about it, except that it is available for very little money. With a few exceptions, OpenAI undercuts its biggest competitors Anthropic, Google, and Mistral AI quite significantly in terms of price. Compared to its top model GPT-4o, the mini version costs just 3% and only about a quarter of what OpenAI charges for the now somewhat outdated GPT-3.5 Turbo, which is already more than a year old.
Here is the current price comparison between the Claude models from Anthropic, the models from Mistral AI from Paris, the Gemini models from Google and finally the models from the German startup Aleph Alpha:
| Input per million tokens | Output per million tokens | ContextWindow | |
| OpenAI GPT-4o mini |
$0.15 | $0.60 | 128K |
| OpenAI GPT-4o |
5$ | 15$ | 128K |
| OpenAI GPT-3.5 Turbo |
$0.50 | $1.50 | 16K |
| Anthropic Claude 3.5 Sonnet |
3 $ | 15$ | 200K |
| Anthropic Claude 3 Opus |
15$ | $75 | 200K |
| Anthropic Claude 3 Haiku |
$0.25 | 1.25 $ | 200K |
| Mistral AI Large |
4 $ | 12 $ | 32K |
| Mistral AI Medium |
$2.7 | $8.1 | 32K |
| Mistral Mixtral 8x22B |
2 $ | 6 $ | 64K |
| Mistral Mixtral 8x7B |
$0.70 | $0.70 | 32K |
| Google Gemini 1.5 Pro |
$3.50
7 $ |
$10.50
21 $ |
128K
> 128K |
| Google Gemini 1.5 Flash |
$0.35
$0.70 |
$1.05
$2.10 |
128K
> 128K |
| Google Gemini 1.0 Pro |
$0.50 | $1.50 | n/a |
| Aleph Alpha Base Control |
7,50€ | n/a | 2,048 |
| Aleph Alpha Extended Control |
11,25€ | n/a | 2,048 |
| Aleph Alpha Supreme Control |
43,75€ | n/a | 2,048 |
The top 3 AI players undercut each other
Here you can clearly see how difficult it is for smaller providers to compete with the big ones in terms of price. Anthropic charges three times as much for its top model “Claude 3 Opus” as OpenAI charges for its flagship model GPT-40, while Google has positioned itself in the middle of the two AI startups with its best LLM.
Most recently, Anthropic launched Claude 3.5 Sonnet and Google launched Gemini 1.5 Flash, versions of their top models with significantly reduced prices – clearly with the aim of luring as many developers as possible away from OpenAi.
Mistral AI from Paris remains well below the prices of the top models from Anthropic, Google, and OpenAI, but cannot keep up with them either, and is actually a bit behind the top 3 in the ranking of the important Chatbot Arena.
And then there is the case of Aleph Alpha: The German startup cannot keep up with its Luminous models in terms of quality (see also context window), but has very expensive models in terms of pricing. What is interesting is that the Aleph Alpha pricing page has also been taken offline, but can still be found here. It can be assumed that Aleph Alpha will soon have to come onto the market with new models and, above all, new prices in order to remain competitive.
