GPT-5编码能力两极分化：效率跃升与预期落差并存
Jenny Wang, engineer and creator of Alta, reports GPT-5 excels at complex coding tasks in a single iteration, citing an example where it successfully generated a press page for her company's website, matching existing design aesthetics. This contrasts with previous models requiring prompt revisions. However, she noted an instance of hallucinated URLs. Another anonymous developer highlighted GPT-5's proficiency in resolving deep technical challenges, mentioning its valuable recommendations and realistic timeline for a network analysis tool project. Enterprise partners like Notion have publicly endorsed GPT-5, with Notion stating it handles complex work 15% better than tested models.
Conversely, some developers express disappointment, deeming GPT-5's coding capabilities to be behind current state-of-the-art expectations. Kieran Klaassen compares its performance to Anthropic's Sonnet 3.5, released earlier. Amir Salihefendić found GPT-5 underwhelming, particularly in coding, likening its release to Meta's Llama 4 disappointment. Mckay Wrigley prefers Claude Code + Opus for coding tasks, labeling GPT-5 as a strong general chat model. Concerns also arise regarding GPT-5's verbosity and redundancy, though this is reportedly adjustable by users.
GPT-5 demonstrates advanced capabilities in complex coding and deep technical problem-solving, as validated by early users and enterprise partners. Its ability to execute intricate tasks in one go signifies a leap in efficiency for developers. However, a segment of the developer community perceives its coding performance as lagging behind current expectations, drawing comparisons to earlier models. This divergence in user experience suggests potential fine-tuning needs or differing application strengths, impacting its reception as a 'state-of-the-art' solution. The model's verbosity, while adjustable, also presents a user experience challenge that requires careful management.
Lauren Goode
网友讨论