More Testing of GPT5 and Comparing Against Other Models

There are various Youtubers on AI who are giving their opinions of OpenAI GPT5. Theo is crowning GPT5 as the best model but others feel it is a good and fast model but are not blown away by it.

GPT5 is the number one model per the LMArena leaderboard with a 1481 score and 1460 for Gemini and 1429 for Grok 4. There are other OpenAI models on the board but the other OpenAI models were deprecated.

There were a variety of tests that Grok 4 did well like simulating smoke or finding Waldo which GPT5 failed.

There in an assessment of what OpenAI GPT5 means for work, research and for coding. GTP5 seems to be prioritizing performance in Canvas. It did a tourist planning task in Canvas using GPT5 but failed for Luvable using GPT5. Checkpoint your coding because applets created by GPT5 can work but are also fragile.

Prompting matters for GPT5 performance. If correctness matters needs think hard in the prompt or think hard button. It then achieves better performance. GPT5 vanilla and pro underperformed even old OpenAI models and other models.

It is a step forward and is progress. It is jump on the coding side and reliability but there is still a lot of work to be done for worldchanging capability for AI.
It is advancing the models jaggedly.
It is advancing what the other models did but a bit better in several cases.

Wes Roth shows significant GPT5 capabilities like a very good Minecraft one shot.