Source URL: https://slashdot.org/story/25/04/14/1726250/openai-unveils-coding-focused-gpt-41-while-phasing-out-gpt-45
Source: Slashdot
Title: OpenAI Unveils Coding-Focused GPT-4.1 While Phasing Out GPT-4.5
Feedly Summary:
AI Summary and Description: Yes
Summary: OpenAI’s launch of the GPT-4.1 model family emphasizes enhanced coding capabilities and instruction adherence. The new models expand token context significantly and introduce a tiered pricing strategy, offering a more cost-effective alternative while providing competitive performance metrics against other leading AI systems.
Detailed Description: OpenAI has announced the release of its GPT-4.1 model family, which includes several variants aimed at improving coding capabilities and the ability to follow instructions. This development is particularly noteworthy for security and compliance professionals working in AI and cloud environments due to its implications for software development and AI usage.
– **Model Variants**:
– Standard GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano
– Available via API, excluding ChatGPT usage
– **Performance Metrics**:
– Scores 54.6% on the SWE-bench Verified software engineering benchmark
– Competitors such as Google’s Gemini 2.5 Pro and Anthropic’s Claude 3.7 Sonnet have higher scores (63.8% and 62.3%, respectively)
– GPT-4.1 achieves 72% accuracy in long video comprehension tests, an improvement from the previous GPT-4o’s 65.3% accuracy
– **Token Context Expansion**:
– Context windows have been expanded to 1 million tokens, translating to around 750,000 words, significantly enhancing its contextual understanding capabilities.
– **Retirement of Older Model**:
– OpenAI will retire GPT-4.5, one of its largest models released recently, effective July 14, indicating a strategy focused on optimizing model performance while reducing costs.
– **Cost Structuring**:
– Pricing is set at $2 per million input tokens and $8 for output tokens for the standard GPT-4.1 model.
– The nano variant is positioned as the “cheapest and fastest model ever,” priced at $0.10 per million input tokens.
– **Knowledge Cutoff**:
– All models incorporated a June 2024 knowledge cutoff, suggesting they can utilize more recent data for contextual relevance compared to previous versions.
The implications of the GPT-4.1 release are significant for professionals focused on AI security and compliance as they consider the integration of such advanced models into their systems, ensuring secure coding practices, and managing the associated risks with the deployment of AI technologies in production environments.