In this new research paper the leading climate models turn out to be either too inaccurate (higher sensitivity) or unalarming (lower sensitivity).
– – –
Plain Language Summary
The last-generation Coupled Model Intercomparison Projects (CMIP6) global circulation models (GCMs) are used by scientists and policymakers to interpret past and future climatic changes and to determine appropriate (adaptation or mitigation) policies to optimally address scenario-related climate-change hazards. However, these models are affected by large uncertainties. For example, their equilibrium climate sensitivity (ECS) varies from 1.83°C to 5.67°C, which makes their 21st-century predicted warming levels very uncertain. This issue is here addressed by testing the GCMs’ global and local performance in predicting the 1980–2021 warming rates against the ERA5-T2m records and by grouping them into three equilibrium climate sensitivity (ECS) classes (low-ECS, 1.80–3.00°C; medium-ECS, 3.01–4.50°C; high-ECS, 4.51–6.00°C). We found that: (a) all models with ECS > 3.0°C overestimate the observed global surface warming; (b) Student t-tests show model failure over 60% (low-ECS) to 81% (high-ECS) of the Earth’s surface. Thus, the high and medium-ECS GCMs do not appear to be consistent with the observations and should not be used for implementing policies based on their scenario forecasts. The low-ECS GCMs perform better, although not optimally; however, they are also found unalarming because for the next decades they predict moderate warming: ΔTpreindustrial→2050 ≲ 2°C.
Tallbloke's Talkshop 
See also Andy May’s recent blog post:
Climate Model Democracy
In the computer modeling world, a world I worked in for 42 years, choosing one model, that matches observations best, is normal best practice. I have not seen a good explanation for why CMIP5 and CMIP6 produce ensemble model means. It seems to be a political solution to a scientific problem. This is addressed in AR6 in Chapter 1,[1] where they refer to averaging multiple models, without considering their accuracy or mutual independence, as “model democracy.” It is unclear if they are being sarcastic.
From the link:
“I do not believe in the collective wisdom of individual ignorance.” – Thomas Carlyle
The average of junk is junk. Ensemble models are junk.
You know their teams have people smart enough to know this obvious truth. Ignoring it suggests corruption.
I’m with Carlyle. When you’re trying to see clearly what’s in a liquid solution, you don’t reduce the transparency with particulate mud. Averaging guesses does not reduce to fact.
Andy May puts it like this…
The model results shown in Figures 1 and 2 resemble a plate of spaghetti. Natural climate variability is cyclical,[10] so this odd practice of averaging multiple models erroneously makes it appear nature plays a small role in climate. Once you average out nature, you manufacture a large climate sensitivity to CO2 or any other factor you wish, and erroneously assign nearly all observed warming to human activities.
. . .
In summary, if the IPCC cannot choose one best model to use to forecast future climate, it is an admission that they do not know what drives climate.
– – –
Manufactured climate sensitivity to CO2 is leading the world up the garden path, as evidenced by the absurdity of labelling it a pollutant.
The modellers have been completely corrupted and any not actually Zealots now dare not speak up. The actual Zealots desperately want ECS to be an emergent property of the models but it is baked in to the models, even if not directly. That I suspect is why some run far too hot – they are double or triple counting ECS through bad modelling.
I note that the Scafetta paper uses a model-friendly temperature dataset ERA5. The graph shows that ERA5 misses the 1998 El Nino warming, and is cooler in the 1980s and 90s, and warmer since 2005-6, compared to UAH6.
Sorry for an error in the difference label. Here is corrected image
Whoops, delete that one was the same.