إطلاق GPT-5: محرك OpenAI الجديد لقوة الترميز ومهام الوكيل

OpenAI has officially released GPT-5 via its API platform, marking a significant leap forward in AI capabilities, particularly for coding and agentic tasks. This new model boasts state-of-the-art performance across key coding benchmarks, surpassing its predecessors in both efficiency and accuracy. GPT-5 is designed to be a true coding collaborator, excelling at code generation, bug fixing, code editing, and complex codebase comprehension. Beyond coding, GPT-5 demonstrates exceptional performance in long-running agentic tasks, showcasing improved tool intelligence and the ability to chain together numerous tool calls reliably. This blog post delves into the key features, improvements, and performance benchmarks of GPT-5, providing developers with a comprehensive overview of this powerful new tool.

Unmatched Coding Prowess

• SOTA Performance: GPT-5 achieves a 74.9% score on SWE-bench Verified and 88% on Aider polyglot, significantly outperforming previous models like o3.
• Efficiency: It achieves these high scores with 22% fewer output tokens and 45% fewer tool calls compared to o3 at high reasoning effort.
• Frontend Excellence: Internal testing reveals that GPT-5 outperforms o3 in frontend web development 70% of the time.
• Real-World Testing: Early testers from companies like Cursor, Windsurf, and Vercel praise its intelligence, steerability, and performance on real-world coding tasks.

Agentic Task Mastery

• Long-Running Tasks: GPT-5 achieves SOTA results on τ2-bench telecom (96.7%), a challenging tool-calling benchmark.
• Tool Intelligence: Improved tool intelligence allows GPT-5 to reliably chain together dozens of tool calls, even in parallel.
• Error Handling: It excels at following tool instructions, handling tool errors, and long-context content retrieval.
• Industry Validation: Companies like Manus, Notion, and Inditex highlight its performance, rapid responses, and depth of reasoning.

New API Features for Enhanced Control

• Verbosity Parameter: Developers can now control the length and comprehensiveness of model responses with the new verbosity parameter (low, medium, high).
• Reasoning Effort Parameter: A minimal value for the reasoning_effort parameter allows for faster answers with less extensive reasoning.
• Custom Tools: A new tool type, custom tools, enables GPT-5 to call tools with plaintext instead of JSON, simplifying complex tool integrations.

Model Sizes and Availability

• Three Sizes: GPT-5 is available in three sizes in the API: gpt-5, gpt-5-mini, and gpt-5-nano, offering flexibility in trading off performance, cost, and latency.
• ChatGPT Integration: While GPT-5 in ChatGPT utilizes a system of reasoning, non-reasoning, and router models, the API version focuses on the reasoning model for maximum performance.
• Non-Reasoning Model: The non-reasoning model used in ChatGPT is available as gpt-5-chat-latest in the API.

Improved Safety and Factuality

• Reduced Hallucinations: GPT-5 makes ~80% fewer factual errors than o3 on prompts from LongFact and FactScore benchmarks.
• Self-Awareness: The model has been trained to be more self-aware of its limitations and better able to handle unexpected situations.
• Health Question Accuracy: GPT-5 demonstrates improved accuracy on health-related questions.

Pricing and Access

• API Availability: GPT-5, GPT-5-mini, and GPT-5-nano are available now in the API platform, supporting the Responses API, Chat Completions API, and Codex CLI.
• Pricing Structure: Pricing varies by model size, with GPT-5 priced at $1.25/1M input tokens and $10/1M output tokens.
• Microsoft Integration: GPT-5 is also launching across Microsoft platforms, including Microsoft 365 Copilot, Copilot, GitHub Copilot, and Azure AI Foundry.

In conclusion, GPT-5 represents a major advancement in AI capabilities, offering developers a powerful tool for coding and agentic tasks. Its superior performance, enhanced control features, and improved safety make it a valuable asset for building innovative applications. The availability of different model sizes and comprehensive API support ensures that developers can tailor GPT-5 to their specific needs and requirements. As GPT-5 is integrated into various platforms, it is poised to revolutionize the way developers approach coding and task automation.

Source: OpenAI