Please don't make the upcoming releases be agentic coding onetrick gimnick models, like every other one so far after the 4.5 series
A user expresses dissatisfaction with recent AI models, stating that models after the 4.5 series have not improved for coding tasks like writing Tampermonkey scripts. They praise the 4.5 lineup, particularly 4.5 Sonnet and 4.5 Opus, for their distinct and manageable output styles, emotional depth, and goal-oriented responses. The user misses the intuitive, oracle-like interface of older models and fears that new releases are becoming "agentic coding onetrick gimnick models," lacking character and consistency, similar to later generations of Beyblades.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 12, 2026, 00:05 UTC
IngestedOffset at this time: UTC+0Sep 12, 2026, 18:01 UTC
- Published
- Sep 12, 2026, 00:05
- Ingested
- Sep 12, 2026, 18:01
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
Register to choose your default translation language in settings.
To this day I have no clue what agentic coding is and I can't force myself to care. For all I have ever done on the coding aspect is asking it to write tampermonkey scripts for myself, it has barely, if at all, improved.
4.5 lineup was my favorite, because the output style was notoriously more managable than ever before or after. 4.5 Sonnet wrote very long texts and 4.5 opus wrote very short. 4.5 sonnet was very emotional while 4.5 opus was goal oriented and didn't waste any word beyond the absolute limit. 4.5 sonnet would figure out connection between couple of abstract concepts, while 4.5 would find inconsistencies and flaws. The extended thinking always worked when enabled and I could diagnosed when something derailed at the exact point, because it wasn't shortened to a point where it says almost nothing.
After that there isn't much to be excited for. The new octopuses have the sonnet writing style, except in more annoying reeking way. User preferences only exist so the bot can figure out how to ignore them. Lowest setting with "will maybe think" toggle turned off can sometimes make it stuck for over a minute, while maxxed out ones with long prompt can get answered in 0.7 seconds (the answer being compltely useless, but atleast it was fast, like in the anecdote.)
I don't care about that the new ones are better at benchmarks, editing excel files, making shitty web designs, hollow unoptimised 3js game prototypes that take 80% of posters RAM to run offline and aren't even fun to play, because why would they, if the only reason is to look somewhat good on paper.
Please give back the intuitive oracle like interface, where the answer wasn't always what I wanted but it indicated that the bot was even trying. If the new ones are closer to AGI™ then I don't want any superintelligence. If it wasn't so sad it would be comical how is it carbon copy rerum of the Beyblades history where every next generation after Metal Fusion were technically superior, but inconsistent and devoid of any character to latch onto so fans were not able to care about them