The GPT-5 model from OpenAI has just undergone updates. The default chat model, GPT-5.1 Instant, has enhancements to its instruction following capabilities. The reasoning model, GPT-5.1 Thinking, is now quicker and produces more intelligible replies. The coding model, known as GPT-5.1-Codex-Max, is trained to employ compaction for extended operations.
The ChatGPT application uses the Instant and Thinking models, which are selected by an auto-routing model. According to OpenAI, users desired models that are fun to converse with, which prompted the firm to alter the default response style of its models. Additionally, the new version allows for greater customization of chat tone and style. The launch includes three new styles, Quirky, Candid, and Professional, in addition to the numerous pre-existing styles in ChatGPT. With personalization settings and unique instructions, users may also alter the response style.
The release of GPT-5 by OpenAI earlier this year sparked a dispute when the option to choose older models was removed, disrupting a large number of user workflows. Despite the fact that ChatGPT still automatically selects the optimum model for a conversation, OpenAI has changed its mind about immediately retiring older models. According to OpenAI:
In the future, when we release new ChatGPT models, our strategy will be to provide users plenty of time to assess the changes and provide feedback, which will enable us to maintain innovating our frontier models while making a seamless transition. With lots of prior notice, sunset times will be clearly communicated.
As part of its attempt to make ChatGPT's personality "[feel] uniquely right" for users, OpenAI updated the model's reaction parameters. Based on "what we’ve learned about how people naturally steer the model," the firm claims that the preset response patterns were created. The fine-tuning options are being made available to a select group of users at a time as an experiment, whereas the presets were made available to all with the release.
The conversational tone of the model failed to wow a number of Hacker News readers, who said they would prefer a more succinct response. One user noted that the "Efficient" (formerly "Robot") preset was designed for this purpose. Another person wrote:
Here, it appears that folks have a rather unfavorable attitude toward a "conversational" AI chatbot. I share everyone else's annoyance at the sycophancy in ChatGPT and its numerous ethical and practical problems, but I don't think that being talkative is a negative trait. I suppose it's simply a matter of taste.
The GPT-5.1-Codex-Max is a better version of OpenAI's GPT-5.1-Codex coding model. When tested on the Verified benchmark of the SWE-bench, GPT-5.1-Codex-Max outperformed the prior model despite utilizing fewer reasoning tokens. In addition to the previous low, medium, and high levels, the model also has an Extra High (xhigh) reasoning effort level.
Benchmark Performance of GPT-5.1-Codex-Max. Image Source: OpenAI Blog
Additionally, GPT-5.1-Codex-Max was the subject of conversation among Hacker News readers.
The Codex CLI is fantastic, and I use it frequently, but I'd want additional useful tools for handling context from CC. I'm really glad that compaction is now available, and I hope that it will provide additional context management capabilities.
One more person stated:
Even if it were behind the "other models" dropdown or something, it would be wonderful to be able to access this model through the chat interface.
The Codex CLI and a few IDE extensions include the new Codex model. "Coming soon" is the availability of API access.
0 Comments