How Effectively Agents Write D

[jmh530]

On Sunday, 8 November 2026 at 23:18:06 UTC, Richard (Rikki) 
Andrew Cattermole wrote:
> On 09/11/2026 3:14 AM, jmh530 wrote:
>> On Sunday, 8 November 2026 at 12:38:48 UTC, Richard (Rikki) 
>> Andrew Cattermole wrote:
>>> [snip]
>>>
>>> I tried that once in the past, and did not care for that user 
>>> experience.
>>> [snip]
>> 
>> How long ago was the past?
>
> 3-4 months
>
> But it was DeepSeek/MiMo.

On Artificial Analysis, Deeseek V4 Pro has a score of 23. Mimo 
2.6 Pro has a score of 30 (this came out in September, so you 
likely used an earlier version, I couldnt easily get 2.5s score, 
or anything that came out more than a few months ago). And that’s 
the reasoning versions of each.

For comparison, GPT-6 Astra medium has an intelligence score of 
52. 6.1 Sol Medium is a 46. 6.1 Terra medium has a score of 38, 
and 6 Luna medium is an 18.

So the intelligence score of the models you tried are somewhere 
between the Luna/Terra level.

At work, if I’m doing something not too difficult in Python and 
Matlab, then I’m ok with 6.1 Sol. It’s a lot cheaper than Astra 
at still capable. But I refuse to put it on Auto, which 
inevitably tries to give me Luna or something. And I spend more 
time figuring out what the hell it’s doing.

So it doesn’t surprise me that you weren’t impressed by the 
models you used. The capability depends on what you are using. 
Not blanket kind of thing.