D
@Dogetothemoon
Grok is getting better everyday
58
1
1.8K
@XFreezeposted on X
The new Grok 4.20 Beta benchmarks are wild 🥇 #1 lowest hallucinating AI (22%) 🥇 #1 at following instructions (83%) 🥈 #2 in agentic tool use (97%) Grok 4.20 ranks #1 in the lowest hallucination rate ever recorded across all AI models tested globally Most models race to sound smart. Grok 4.20 was built to never lie and still dominates on instruction following and agentic tasks This is literally a 500B model performing top-notch in the things that matter most
View original tweet on X →
Community Sentiment Analysis
Real-time analysis of public opinion and engagement
Sentiment Distribution
67% Engaged
39% Positive
28% Negative
Positive
39%
Negative
28%
Neutral
32%
Key Takeaways
What the community is saying — both sides
Supporting
1
500B model with a record-low 22% hallucination rate
built to favor truth over flair, and they celebrate it as the future of reliable AI.
2
reliability
lower hallucinations plus strong instruction‑following translate into enterprise readiness and close the gap between chatbots and “AI workers.”
3
agentic/tool use
(Cursor, Codex‑style capabilities, multi‑agent workflows) and whether those benchmarked scores hold up in production.
4
hallucination avoidance
is important, yet some frontier models still top combined intelligence and reasoning benchmarks.
5
less restrictive content filters
, arguing that freedom of execution (not moral filters) accelerates productivity and innovation.
6
hands‑on deployment feedback
, case studies, and long‑running agentic behavior before changing critical systems.
Opposing
1
cherry-picked marketing stunt
tests can be "chopped", curated, or tuned to look good without proving real capability.
2
22% hallucination
is framed as unacceptable: that's "one-in-five responses wrong" and a clear warning for any high-value or autonomous use.
3
conservatism or abstention
(benchmarks reward saying "I don't know") or strict output validation, not necessarily better understanding.
4
LLMs do not truly reason
they mirror training data and will keep producing confident falsehoods until reasoning is solved differently.
5
regressions across releases
(Grok 4 → 4.1 → 4.
6
and question value when newer builds reintroduce hallucinations or break prior b...
and question value when newer builds reintroduce hallucinations or break prior behavior.
7
throwing half a trillion parameters
or bigger models is seen as an expensive band-aid, not a fundamental fix.
8
usability failures
poor navigation, ignoring important code pages, bad schema/syntax validation, and fragile instruction-following.
9
independent, long-term validation
and steady behavior over time rather than trumpeting one-off benchmark wins.
Top Reactions
Most popular replies, ranked by engagement
M
@Metzes77
🔥 Grok 4.20 Beta is an absolute monster! 🥇 A 500B model built to never lie – and still dominates everything! Truth first wins. xAI rocks the world! 🚀 #Grok @xai
45
5
4.1K
0
@0xMariussi
Lowest hallucination at 22% is still 1 in 5 responses being wrong. That's not a flex, that's a warning label.
27
10
3.3K
S
@SeeBx
he top .5 % users of ChatGPT. Grok is very simlar but it is still overly weighed down by the data it is fed. It has very little ability to reason. But no LLM is able to reason. It will never be anything more than mirror until we solve that problem. They all easily spew very fal
16
7
2.5K
D
@dimkovska88
While other models are racing for 'intelligence' points, Grok is out here actually making sure the answers are right. This is exactly what the industry needs!
14
1
1.2K
K
@kamwbe
Grok at navigation leaves something to be desired. I told it to go home and it took a very round about route that was 5 times as long. From now on I will give it more explicit directions
11
4
1.5K
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.