OpenAI CEO Sam Altman Hails GPT-5.4 as Favorite Model While Acknowledging Three Key Weaknesses
Key Points
- OpenAI CEO Sam Altman calls GPT-5.4 his favorite model for conversation.
- GPT-5.4 shows notable gains in personality and coding ability.
- Three weaknesses are identified: frontend aesthetic taste, real‑world context gaps, and incomplete task execution.
- Altman assures the company will address these issues.
- User feedback highlights competition with models like Claude, Gemini and Opus.
- Community efforts continue to revive the retired ChatGPT‑4o model.
OpenAI chief Sam Altman praised the new GPT-5.4 model as his favorite version to converse with, highlighting improvements in personality and coding ability. He also recognized three shortcomings—frontend aesthetic taste, occasional lapses in real‑world context, and incomplete task execution—that the company plans to address. The remarks underscore OpenAI’s shift toward refining how ChatGPT feels to use, not just its raw performance, as it competes with rivals such as Claude, Gemini and Opus.
Altman’s Enthusiastic Endorsement
In a recent post on X, OpenAI chief executive Sam Altman described GPT-5.4 as “my favorite model to talk to!” He emphasized that the latest fifth‑generation model delivers a more engaging personality and excels at coding, knowledge work, and computer‑based tasks. Altman noted that OpenAI had previously “missed the mark on model personality for awhile,” and that GPT-5.4 represents a step in the right direction.
Identified Weaknesses
Responding to feedback from user Matt Shumer, Altman also listed three areas where GPT-5.4 still falls short. The first is “frontend taste,” which refers to the model’s sense of style and aesthetics when generating user‑interface designs, a dimension where rivals like Opus 4.6, Gemini 3.1 Pro, and Claude are perceived as stronger. The second weakness involves occasional failures to incorporate obvious real‑world context; for example, the model suggested travel itinerary locations that would be overcrowded by spring breakers, requiring a prompt rewrite. The third issue concerns task completion; during testing within the OpenClaw system—a platform for automating AI workloads on Mac clusters—the model stopped short before finishing tasks.
OpenAI’s Response and Outlook
Altman responded positively, stating, “We will be able to fix these three things!” His comments suggest that OpenAI is now prioritizing the user experience of ChatGPT—its tone, personality, and ease of interaction—alongside traditional performance metrics. This focus comes amid ongoing community campaigns to restore the retired ChatGPT‑4o model, which many users felt had a superior personality compared with earlier fifth‑generation releases.
Competitive Landscape
The discussion highlights a broader competitive environment in which OpenAI’s GPT models are measured not only by raw capability but also by aesthetic and contextual fluency. Rivals such as Claude, Gemini, and Opus are cited as benchmarks for frontend design quality, underscoring the importance of visual polish in AI‑generated outputs.
Implications for Users
For developers and knowledge workers, GPT-5.4’s strong coding performance and improved conversational tone are promising. However, the noted weaknesses remind users that careful prompt engineering and verification remain essential, especially for tasks involving real‑world planning or complex automation workflows.