OpenAI's latest AI model, GPT-4.5 (code-named "Orion"), delivers augmented capabilities across multiple domains with improved accuracy rates of 62.5% on SimpleQA benchmarks. You'll find it available through ChatGPT Pro subscriptions and OpenAI's API, with plans to expand access to Plus and Team subscribers. The model comes with premium pricing at $75 per million input tokens and $150 per million output tokens. This release marks a crucial shift in AI development, as the industry moves beyond traditional scaling approaches.
OpenAI's latest breakthrough, GPT-4.5, has arrived as a research preview, marking the company's largest and most sophisticated AI model to date. Code-named "Orion," this new model relies on the same unsupervised learning techniques that powered its predecessors but delivers notably improved capabilities across multiple domains.
GPT-4.5, known as Orion, emerges as OpenAI's most advanced AI model, building on proven techniques while expanding capabilities substantially.
You'll find GPT-4.5 initially accessible through ChatGPT Pro subscriptions and OpenAI's API, with plans to expand access to ChatGPT Plus and Team subscribers next week. The model showcases enhanced conversational abilities and demonstrates a marked reduction in hallucinations compared to GPT-4o. OpenAI specifically designed this release to deliver less prone to hallucinations than any previous version.
The new model's performance metrics reveal a 62.5% accuracy rate on SimpleQA benchmarks, outperforming GPT-4o in pattern recognition tasks. However, it's worth noting that GPT-4.5 still faces challenges in logical reasoning compared to specialized models like Claude 3.7 Sonnet, particularly in coding applications. The model stands as the last non-chain-of-thought system before OpenAI's shift toward reasoning-focused development.
You'll need to take into account the substantial cost implications of using GPT-4.5. The model charges $75 per million input tokens and $150 per million output tokens, making it considerably more expensive than GPT-4o. These higher costs reflect the computational demands of running such a large-scale model.
OpenAI's release of GPT-4.5 signals a broader shift in AI development strategy. The industry's growing recognition of diminishing returns from traditional scaling has prompted a new focus on integrating reasoning capabilities. This change will influence OpenAI's future developments, including their planned GPT-5 release.
For enterprise users, GPT-4.5 access comes with a $200 monthly subscription through ChatGPT Pro. The model's multimodal capabilities support file and image uploads, while its improved world knowledge and emotional intelligence make it particularly valuable for specific industry applications.
This release represents a vital moment in AI development, highlighting both the potential and limitations of scaling traditional language models. As the industry grapples with these challenges, the focus increasingly shifts toward developing more efficient and reasoning-capable AI systems.
Frequently Asked Questions
Can GPT-4.5 Write Code Better Than Human Programmers?
While GPT-4.5 can generate novel code solutions and assist with programming tasks, it doesn't consistently outperform human programmers.
You'll find that experienced developers are still better at complex problem-solving, debugging, and error handling.
GPT-4.5 serves better as a collaborative tool to improve your coding process rather than replace human expertise.
It matches human abilities in some areas but falls short in logic-heavy programming tasks.
How Much Will GPT-4.5 Cost for Individual Users?
You'll need a ChatGPT Pro subscription at $200 per month to access GPT-4.5 currently.
However, it will soon be available to ChatGPT+ users at $20 per month.
If you're a developer using the API, you'll pay $75 per million input tokens and $150 per million output tokens.
There's a reduced rate of $37.50 per million tokens for cached inputs.
These prices are notably higher than previous models, though cheaper options may become available later.
Does GPT-4.5 Require Special Hardware to Run Effectively?
You won't need special hardware to use GPT-4.5 as an end user – it runs on OpenAI's servers, not your local device.
However, OpenAI requires extensive hardware infrastructure, including tens of thousands of GPUs, to operate the model effectively.
The company's computing demands are so substantial that they're investing in custom semiconductors and expanding their GPU capacity to handle the model's increased complexity and processing requirements.
Will GPT-4.5 Be Available in Languages Other Than English?
Yes, you'll be able to use GPT-4.5 in multiple languages.
The model has demonstrated high accuracy in languages including English, Spanish, and Arabic on MMLU benchmarks, making it suitable for global applications.
While the initial rollout may prioritize English, there aren't any announced language-based restrictions.
You can expect broad language support as OpenAI expands access to Plus, Team, Enterprise, and Edu users in the coming weeks.
How Does GPT-4.5's Accuracy Compare to Previous Versions in Medical Diagnoses?
While specific comparative data for GPT-4.5's medical diagnostic accuracy isn't yet available, you'll find it builds on GPT-4's strong performance in healthcare.
GPT-4 achieved 72-100% accuracy in dermatologic conditions and showed expertise comparable to medical residents in radiology.
Current studies suggest GPT-4.5's reduced hallucination rate and improved world knowledge might bolster diagnostic reliability, but you'll need to wait for peer-reviewed research to confirm these potential advantages.
