D
@DuaneStorey
Opus 4.7 won't even do medical research for me anymore, I always get some usage policy exemption, even when I ask it to go read public PDFs on the web. It's mostly useless for me now. I switched to Codex about a week ago.
19
3
760
@GergelyOroszposted on X
I regularly run ChatGPT and Claude side-by side. ChatGPT gets the same task done, while Claude just doesn't... I am souring on Anthropic based on my recent experiences What is the point of an AI tool if I have to manually click 3-4x to have it complete my task... https://t.co/yXUk8A4ErL
View original tweet on X →
Community Sentiment Analysis
Real-time analysis of public opinion and engagement
Sentiment Distribution
83% Engaged
71% Positive
Positive
71%
Negative
12%
Neutral
16%
Key Takeaways
What the community is saying — both sides
Supporting
1
Claude has noticeably degraded
users report it’s become slower, times out mid-task, suffers context rot and “refusal” loops, and performs worse than it did a month ago or in earlier Opus releases.
2
Tool/session limits break workflows
the model hitting a “tool-use ceiling” or session limit mid-task causes repeated restarts, wasted tokens and hidden costs — many call it effectively looting users.
3
Execution and reliability beat peak reasoning
if a model needs babysitting or multiple clicks to finish, it’s useless for work. People are switching to ChatGPT / Codex / OpenCode because they reliably finish one-shot tasks and save time.
4
Training-data behavior explains some regressions
one hypothesis is Opus 4.7 was trained on session-max data, teaching the model to overuse tools, load extra skills and defer work — and that behavior now clashes with Anthropic’s harness limits.
5
Agentic loops vs one-shots
Claude (and newer Opus versions) can excel in long, agentic loops and complex multi-step PRs, while GPT variants win on throughput and single-shot execution; mixing models in a stack is a common recommendation.
6
Enterprise-first tradeoffs hurt UX
many feel Anthropic has prioritized B2B features and safety at the cost of the consumer experience — poor support, strict identity checks and neglected consumer dashboards are recurring complaints.
7
Practical workarounds and lessons
users are switching models, routing tasks to the tool that “just works,” or using CLI flags (e.g. --dangerously-skip-permissions) to avoid stoppages — the pragmatic takeaway: pick reliability over headline capability.
Opposing
1
Claude is on the rise
some predict Anthropic (with Google's Gemini looming) will soon be the best LLM.
2
Real user migration
reports of millions uninstalling ChatGPT and Claude hitting #1 on the App Store are cited as proof of changing market preference.
3
Intentional safety tradeoffs
Claude’s extra clicks and tool limits are framed as deliberate design choices to reduce risky outputs.
4
Non-technical vs technical split
many non-coders praise Claude as “smarter,” while some developers report opposite experiences, revealing a perception gap.
5
Prompting and workflow matter
frequent retries often signal poor framing of the task rather than an inherent model failure.
6
One-shot completion vs cautious stopping
ChatGPT often returns a full (sometimes hallucinated) solution in one pass; Claude stops to avoid hallucination and relies on tools, which can require extra user interaction.
7
Clicks are trivial to some, infuriating to others
defenders call the extra steps benign or even “character building,” while critics accuse complainers of laziness.
8
Hybrid toolchains are common
developers report using one model as a coding engine (Codex/ChatGPT) and Claude for code review or tooling integrations.
9
Anecdote-heavy debate
most claims are presented as personal experience rather than systematic comparison, so opinions remain mixed.
Top Reactions
Most popular replies, ranked by engagement
C
@championswimmer
The consumer experience is not even bad, it is totally non-existent. I am sure no one even looks at the consumer app dashboards internally, they are printing so much money on enterprise. They are truly a B2B SaaS company.
19
2
1.0K
W
@WillToMake
Wait til they take your $200 and THEN ask for 2 forms of government ID or you can't use the services you already paid for @bcherny @_catwu how do I get my $200 back? I'm moving exclusively to Codex
11
0
439
M
@mfreeman451
codex is my daily driver, and I'll have claude code do a review on my PR, adding comments using 'gh'
1
0
51
A
@aalachimo
Honestly this sounds less like “Claude vs ChatGPT” and more like a prompting/workflow issue. If you need 3–4 retries, the task probably isn’t being framed clearly enough for either model to nail it first pass.
1
0
41
G
@GergelyOrosz
web interface
0
1
2.4K
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.