Well am sure some of you have already seen Open AI released the GPT-5 model family including GPT-5, GPT-5 Mini, and GPT-5 Nano. By all accounts, and by the standards of many reputable Developers, it is truly groundbreaking, concerning, and incredible at the same time. We are just about to play with it in Open Router to gain our own insights... Could spell doom for a lot of Junior Developers as can see this latest incarnation removing the need for them all together as it really is that good going by the reports of industry professionals who themselves have had early access for weeks now. All are reporting the same. Groundbreaking and next level.
You can enable the Preview of GPT 5 in your Copilot settings:
But for some reason you can not yet select it so maybe have to wait for a Copilot update to roll-out...?
The latest Cline update also allows selecting GPT 5 models.
1 Like
Have you by any chance tried insiders build? They usually have beta/alpha copilot updates
Excited to test it on Warp 
1 Like
No I've not tried that yet. I was going to see if the Pre-Release version has it available. Quite happy to use Open Router though. Pricing is reasonable which was a pleasant surprise. Actually cheaper than Claude by five Dollars on the output side. Will also apply the same multiplier (x1) as other premium models in Copilot (again surprised). Open AI definitely needed to pull something out of the bag. Be interesting to see how the other providers respond in the next few weeks.
Quite frankly, using OPUS for Wappler-related bindings is about as useful as Sonnet, but it chews through tokens as if you were a Nigerian prince with millions in inheritance. I dislike Grok, so I’m glad OpenAI now offers a coding model that is as good of nott better than anthropic flagship one and isn’t priced at £100 or a million-token output.
In general, I find it fascinating how competitive the market is and how the next model could shake things up, given the tight margins these companies—at least those in Silicon Valley, not the Chinese ones trained on synthetic ChatGPT data—are running. 
2 Likes
1 Like
Had a play at code level, few breaking changes
Obvious ones
'temparature' is no longer supported although it appears it will accept a value of 1 otherwise it thrown an error
'max_tokens' becomes 'max_completion_tokens', again error is thrown
New options include:
reasoning_effort (low,medium,high)
verbosity (low,medium,high)
Busy re-engineering "All-in-one-AI" extension to deal with those changes
EDIT: "top_p" ( determinism) and "presence_penalty" (deals with repeating text) also seem to have been dropped.
3 Likes
I just tested it and it shows in the Wappler dropdown.
1 Like
Has anyone used GPT-5 to create a Summernote replacement? 
6 Likes
That’s the HOLY GRAIL 
2 Likes
Perhaps an alternative would be to identify what changes are needed and have ai fork it to extend or amend the current functionality.
1 Like
I’m not impressed by GPT-5, it doesn’t live up to the hype. I don’t see a major improvement, except for the fact that they’ve now merged all the models into one, which I think is a great move. But in terms of capabilities, it leaves a lot to be desired. I’ve even seen Gemini Pro perform better at coding tasks than GPT-5.
That said, it’s still a bit early to judge since I haven’t tested it extensively yet. Here’s a comparison video I thought was really good:
1 Like
Here the Code Report about Gpt-5 from Fireship XD
2 Likes
@Max_Saravia I was waiting for the Fireship video! If you know you know...
1 Like
From our initial tests with GPT 5 can't say am that impressed. On the U.I front is a little better and more modern but still a GPT model and to be honest not really meeting the initial hype for us. Yes it can do some cool stuff with 3D but unless you're using 3D in your application then its pretty much redundant. Still finding Gemini 2.5 Pro, Deepseek and Qwen, are working the best for us (for JavaScript which is our primary language). Quite like Kimi K2 as well after playing around with it. I'd prefer to use Claude over GPT if I was pushed to make a choice between the two otherwise we are going to stick with what works for us. All in all GPT 5 was a bit disappointing after all the hype (lot of fanbois out there who all seem to have been spawned by Y Combinator start-ups and we all know who is the poster boy for them). I'm not sure what it is that the GPT models are missing...? Apparently it is very good at identifying skateboard tricks though!
I'll say this though. We shall try again in a few days as am sure there will be many enhancements made to GPT 5 so won't entirely write it off just yet!
I've actually... sort of... grown to appreciate it!
My current issue with exclusively using Claude is that instead of providing solutions to my specific problem, it ALWAYS returns alternative solutions! For example, instead of using custom ParadeDB (with the pg_search @@@ operator that essentially integrates ETL within the PostgreSQL database itself and delivers Elasticsearch-level performance directly from PostgreSQL), it falls back to ILIKE when it encounters the same problem multiple times. Even when EXPLICITLY told not to use any operators other than @@@ for search, it proceeds to implement alternatives just to return SOMETHING. It's driving me crazy.
GPT-5, although slower, has never once tried to use operators other than @@@. The same applies when writing Elixir functions for connecting normal PostgreSQL tables to Typesense via Sequin pipeline. While GPT-5 sometimes provides incorrect structure or outright wrong comment syntax that breaks the function, Claude went so far as to create a separate dockerized worker instance that was supposed to transform the data before it even reached the Sequin pipeline(?). I was shocked by this choice, given that Sequin is built for exactly this use case! Please note: they were given the exact same prompts!
It hasn't been a full week since the rollout, but it's been OKAY in my workflow. It certainly isn't the humanity-ending LLM that many YouTubers made it out to be! However, for Elixir and SQL functions so far, it's been decent. I'm looking forward to testing it with backend and frontend in Wappler now that my etl workflows are finished. And who knows, maybe I'll create a cheeky Tony Hawk fan page since it's supposedly that good at skateboard tricks! 
2 Likes