Horizon Beta (ChatGPT 5?)

I am personally still doubtful that this is a new frontier model from OpenAI. My suspicion remains that this is Deepseek V4, though this is purely based on a mix of pure feelings, the speed (slightly higher than V3 was at launch, far higher than it is from Deepseek directly now; could potentially line up with them using locally sourced accelerators over Nvidia now), the timeline, size and tokenizer. Would be very impressive if it was. Horizon Beta does not perform markedly better over GPT-4.1, some lauded aspects such as the purported frontend proficiency do not translate amazingly well to longer term development [0], so if Horizon Beta is GPT-5 that would be disappointing to me personally, especially considering Horizon Beta does very poorly on tool call and MCP evals in my scenarios, making it less suitable for Agentic coding tasks. In that area, it is even worse than Gemini 2.5 Pro which I have reliably seen end up in continuous loops when failing test cases.

[0] Basically, yes, one shot Horizon Beta outputs "more" UI (very expansive mockups), but the second one uses it to improve interface sections in an existing code base, Horizon Beta is roughly equivalent to Sonnet, GPT-4.1, K2 and 2.5 Pro. Whether a dev wants their initial prompt to create an extensive interface is honestly more a question of preference over model training or performance. Some will like it, some will find it restrictive. In either case, similarly extensive one shot UI code can be achieved with e.g. prompting GPT-4.1 if one wants that.