Skip to content
The Chat Archive
sonar-pro LMArena Search Arena 24K

gpt 4.5

  1. User

    gpt 4.5

  2. Assistant

    GPT-4.5 is OpenAI's latest and most advanced language model, released as a research preview on February 27, 2025. Here are the key points about GPT-4.5:

    Capabilities and Improvements

    • Largest and most capable GPT model for chat to date[1][2]
    • Enhanced natural conversation and emotional intelligence[1]
    • Broader knowledge base and improved ability to follow user intent[2]
    • Better at creative and nuanced tasks like writing and problem-solving[2]
    • Expected to hallucinate less than previous models[2]
    • Supports content generation, multilingual proficiency, and programming assistance[1]

    Technical Details

    • Significantly larger than previous models, estimated to be 5-7T parameters total[3]
    • Focused on advancing unsupervised learning rather than chain-of-thought reasoning[1]
    • Improved performance on various benchmarks, including GPQA, PersonQA, and multilingual evaluations[3]

    Availability and Integration

    • Available to Pro, Plus, Team, Enterprise, and Edu users of ChatGPT[2]
    • Accessible via web, mobile, and desktop platforms[2]
    • Integrated with ChatGPT features like file/image uploads, canvas, and search[4]
    • Available to developers on paid API tiers with support for key features[5]

    Limitations and Concerns

    • Lacks chain-of-thought reasoning capabilities[2]
    • Can be slower due to its size[2]
    • More expensive than previous models, raising concerns about practicality for some users[6]
    • Still experiences hallucinations, with reports of fabricating information 37% of the time in certain tests[6]

    Reception and Controversy

    • Mixed reactions from the AI community, with some questioning the focus on emotional intelligence over analytical power[6]
    • Criticism of OpenAI's "vibes-based" benchmarking approach instead of traditional, objective measurements[6]
    • Concerns about the high cost compared to competitive models from other companies[6]

    GPT-4.5 represents a significant step in scaling up unsupervised learning, but its release has also sparked debates about the direction of AI development and OpenAI's priorities.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.