Big Data Digest is reproduced with permission from Xi Xiaoyao Technology Talks Author | Wang Ergou
GPT-4.5 Information is suspected to be leaked!
According to leaked information, the pricing of this OpenAI's most advanced GPT 4.5 model has been exposed, and the multi-modal function has been greatly upgraded - supporting multi-modal functions across languages, audio, vision, video and 3D, as well as complex reasoning and cross-modality State understanding ability.
The earliest reddit post leaking screenshots of this information has been deleted.
As can be seen from the leaked screenshots, OpenAI has launched three models this time:
• GPT-4.5, $0.06 per 1,000 input tokens, $0.18 per 1,000 output tokens;
• GPT-4.5-64k, $0.18 per 1,000 output tokens; USD 0.12 per thousand input tokens, USD 0.36 per thousand output tokens;
• GPT-4.5-Audio & Voice, USD 0.012 per minute input, USD 0.024 per minute output;
It can be seen that compared with the previous GPT-4 Turbo model , the price of GPT-4.5 has increased by a full 6 times, and the price of GPT-4.5-64k has increased by 12 times. The price increase of
must give people great expectations. Is the performance improvement worthy of this price?
Although the leak of GPT 4.5 has not been confirmed, as early as a week ago, according to "reliable" news from Twitter user @apples_jimmy: "OpenAI may release GPT-4.5 before the end of December."
is well-known in the technology circle The person who broke the news, futuristflower, also believes that the information leaked by the screenshots is basically correct, but it is impossible to verify whether the screenshots are official.
In addition, according to the latest news from Sam Altman on Twitter today, ChatGPT Plus has reopened for registration and subscription. More importantly, Sam Altman claimed that OpenAI has found more GPUs.
Some netizens asked whether this means that (GPT-5) training is over?
Because previously, Sam Altman admitted that GPT-5 was under development.
Perhaps GPT-5 is still “on the way”, but GPT 4.5 may indeed have arrived.
This time GPT 4.5 may be able to completely defeat the Google Gemini Utral model, and I will be convinced.
Because Microsoft countered Google two days ago, it was misleading when releasing information. Gemini Ultra uses a more complex Prompt form. If only standard Prompt is used, Gemini Ultra's performance will be worse than GPT-4.
And if GPT-4 uses the Medprompt method that Microsoft just proposed two days ago, it will defeat Gemini and get a new SOTA:
Will GPT-4.5 still be lazy?
In addition, when they learned that GPT 4.5 was suspected to be leaked, the first thing that many netizens thought of was to complain about the performance degradation of GPT-4.
Recently, ChatGPT’s “becoming lazy” has become a hot topic, and ChatGPT officials also issued a document indirectly admitting that ChatGPT has become lazy and its performance has declined.
Some related research even believes that ChatGPT has learned to "take winter vacation" like humans.
Of course, there is a solution to ChatGPT becoming lazy. Some experts directly came up with an extended Plus summary version of the instructions to avoid GPT becoming lazy for your reference:
-
-
-
-
-
-
-
-
-
-