- Performance assessment
- # More stable: 4/4 reliability standards
- ** Shows good against mainstream models**
- Tool integration: not only language models, but also intelligent assistants
- **o3 prices down 80% below GPT 4o **
OpenAI officially launched a new generation model, O3-Pro, which is now open to all ChatGPT professional and API users and will soon be extended to the corporate and educational versions. Compared to previous versions (e.g. o3 and o1-Pro), O3-Pro has achieved significant improvements in a number of ways. The expert review highly evaluated its capabilities in the fields of science, education, data analysis, writing and programming, and noted that it is more excellent in terms of clarity, integrity of content, compliance with directives and accuracy.
-
** The expert assessment indicates** that performance is better than that of o3,o3-pro in a number of key areas, including science, education, programming, data analysis and writing.
-
O3-Pro rated higher ** clarity, comprehensiveness, command compliance and accuracy**.
-
Similar to O1-Pro, o3-Pro, excellent in mathematics, science and programming in academic assessment.
-
OpenAI’s “4/4 Reliability Assessment” was used to test model stability - only if all of the four responses were correct will it be successful.
-
O3-Pro has access to ChatGPT advanced tools: Web search, file analysis, image recognition, Python programming, personalization of memory, etc.
-
o3-Pro to become a default model for professional and team-based users from this very day on, replacing o1-pro; users of enterprise and educational versions will be granted access next week.
Performance assessment
In expert assessments, evaluators generally prefer o3-pro to o3, emphasizing performance enhancement in key areas such as science, education, programming, data analysis and writing.
O3-Pro obtained higher scores in terms of clarity, comprehensiveness, command understanding and execution, accuracy of content, etc.
# More stable: 4/4 reliability standards
In order to verify the stability of the model, OpenAI uses a rigorous evaluation criterion called “4/4 response” (all answers are correct in four). Only if the model is able to answer the question correctly in four consecutive attempts is it considered to be truly reliable.
In this assessment, O3-Pro’s performance proved not only that it was smart, but that it was stable.
** Shows good against mainstream models**
-
Performance is comparable to Gemini 2.5 Pro and is consistent with the intelligence index in the evaluation
-
Smart is better than Claude 4 Sonnet Tinking, but the unit cost is lower
** Less concise than Claude 4 Opus** but ** than Gemini 2.5 Pro and DeepSeek R1**
Tool integration: not only language models, but also intelligent assistants
O3-pro is much more than “chat chat” AI, which brings together all the high-level tools of ChatGPT, making it a truly multi-skilled intellectual assistant:
-
Web search: quick search and integration of real-time information
-
Document analysis**: available for reading and understanding the contents of the uploaded document
-
Image recognition**: capable of processing visual input and understanding images
-
Python programming capability: direct operation code, data processing, drawing, etc.
-
Personalized memory**: Keeping in mind user preferences and continuously optimizing interactive experiences
**o3 prices down 80% below GPT 4o **
- OpenAI substantially reduced the price of the o3 model 80%: from $8.40 per million input/output token to $2/ $8, while providing a **75% discount for cache input token. o3Pro price: input: $2/ 1 million
o1-Pro: $600
o3-Pro: $20
O3-Pro is better than O1-Pro in every way, faster, smarter and stronger.
And it’s 30 times cheaper.
- o3 The cost per token is equal to GPT-4.1: The price is uniform between reasoning and non-extremistic models, but the actual cost per request is still higher due to o3 **the average output is about 7 times the GPT-4.1 **.
O3 unit token cost aligned to GPT-4.1
-
The two are at the same price as “each token”.
-
But it’s a lot lower than GPT 4o.
-
But since the number of tokens generated by o3 ** is about 7 times as high as GPT-4.1 , the cost of a full query is higher**
-
GPT-4 level smarts have declined since their release **100 times more **
-
Trained models that meet the “advanced intelligence threshold” and whose economic thresholds continue to fall
-
The marginal cost of the AI service continues to decline for the user deployment, contributing to more landing scenarios
This means that the higher-performance AI model is becoming more usable and better suited for commercial integration and large-scale use.
