/Tech12h ago

LisanBench creator @scaling01 finds Anthropic averaged a 6.63 sentiment score compared to OpenAI's 4.89 across 4,000 tweets

OpenAI sentiment declined steadily starting with GPT-5.

39401136339.3K

#349

Original post

Lisan al Gaib@scaling01#1185inTech

I analyzed the sentiment of every tweet I’ve posted about OpenAI and Anthropic.

I had Codex score ~4,000 company mentions from 0-10, where 0 is extremely negative and 10 is extremely positive.

Average scores: OpenAI: 4.89 Anthropic: 6.63

For Anthropic, my most negative stretch was between Claude 4 and Claude 4.1. There’s a clear uptick around Opus 4.5, then a slight uplift by Mythos and then massive crash with Opus 4.7.

For OpenAI, sentiment starts sliding hard from GPT-5 through GPT-5.3, probably because of the token-maxxing era. There’s also a sharp crash right before GPT-5.4, likely the supply-chain-risk backstabbing episode.

GPT-5.5 looks like the most positive OpenAI launch in the dataset so far.

6:04 AM · Jun 6, 2026 · 16.8K Views

/Tech12h ago

LisanBench creator @scaling01 finds Anthropic averaged a 6.63 sentiment score compared to OpenAI's 4.89 across 4,000 tweets

OpenAI sentiment declined steadily starting with GPT-5.

39401136339.3K

#349

Original post

Lisan al Gaib@scaling01#1185inTech

I analyzed the sentiment of every tweet I’ve posted about OpenAI and Anthropic.

I had Codex score ~4,000 company mentions from 0-10, where 0 is extremely negative and 10 is extremely positive.

Average scores: OpenAI: 4.89 Anthropic: 6.63

For Anthropic, my most negative stretch was between Claude 4 and Claude 4.1. There’s a clear uptick around Opus 4.5, then a slight uplift by Mythos and then massive crash with Opus 4.7.

GPT-5.5 looks like the most positive OpenAI launch in the dataset so far.

6:04 AM · Jun 6, 2026 · 16.8K Views

Sentiment

Positive users call technical analyses of OpenAI-Anthropic sentiment thoughtful and accurate, while negative users dismiss them as astrology or reply with personal insults.

Pos

50.0%

Neg

50.0%

8 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

VIEWS10.7KBOOKMARKS13LIKES94REPLIES9

Lisan al Gaib@scaling01

With the current slope of my sentiment since the launch of GPT-5.3-Codex, I will turn more bullish on OpenAI than Anthropic on September 30, 2026

(incredibly realistic forecast)

4h10.7K9413

RETWEETS3

Lisan al Gaib@scaling01

if you are not doing technical analysis on your own sentiment then what are you even doing

Lisan al Gaib@scaling01

With the current slope of my sentiment since the launch of GPT-5.3-Codex, I will turn more bullish on OpenAI than Anthropic on September 30, 2026

(incredibly realistic forecast)

4h6.1K6511

Lisan al Gaib@scaling01

even image gen is bullish on Anthropic

Lisan al Gaib@scaling01

if you are not doing technical analysis on your own sentiment then what are you even doing

4h1.7K92

jason@jxnlco

@scaling01 Can you plot the relative ratio between them over time

Lisan al Gaib@scaling01

I analyzed the sentiment of every tweet I’ve posted about OpenAI and Anthropic.

I had Codex score ~4,000 company mentions from 0-10, where 0 is extremely negative and 10 is extremely positive.

Average scores: OpenAI: 4.89 Anthropic: 6.63

For Anthropic, my most negative stretch was between Claude 4 and Claude 4.1. There’s a clear uptick around Opus 4.5, then a slight uplift by Mythos and then massive crash with Opus 4.7.

GPT-5.5 looks like the most positive OpenAI launch in the dataset so far.

6h1.4K81

Gary Marcus@GaryMarcus

@scaling01 it’s not bias; OpenAI deserves the more negative attention because of their long track record.

Lisan al Gaib@scaling01

I analyzed the sentiment of every tweet I’ve posted about OpenAI and Anthropic.

I had Codex score ~4,000 company mentions from 0-10, where 0 is extremely negative and 10 is extremely positive.

Average scores: OpenAI: 4.89 Anthropic: 6.63

For Anthropic, my most negative stretch was between Claude 4 and Claude 4.1. There’s a clear uptick around Opus 4.5, then a slight uplift by Mythos and then massive crash with Opus 4.7.

GPT-5.5 looks like the most positive OpenAI launch in the dataset so far.

3h33670

Lisan al Gaib@scaling01

@jxnlco same plot just different smoothing

jason@jxnlco

@scaling01 Can you plot the relative ratio between them over time

6h61531

Lisan al Gaib@scaling01

i should have also included notable dates for political drama

Lisan al Gaib@scaling01

I analyzed the sentiment of every tweet I’ve posted about OpenAI and Anthropic.

I had Codex score ~4,000 company mentions from 0-10, where 0 is extremely negative and 10 is extremely positive.

Average scores: OpenAI: 4.89 Anthropic: 6.63

For Anthropic, my most negative stretch was between Claude 4 and Claude 4.1. There’s a clear uptick around Opus 4.5, then a slight uplift by Mythos and then massive crash with Opus 4.7.

GPT-5.5 looks like the most positive OpenAI launch in the dataset so far.

7h1.9K50

A fierce pancake@SayItLoud19

@scaling01 You need to get out more.

12h1861

μck ٤:@JustMicrock

@scaling01 how did you do it? just fed all your posts to an agent?

6h131

haro@harobuilds

@scaling01 the ratio has been compressing since codex launched but "september 30" is doing a lot of heavy lifting for a metric that swings 0.2x on a single release cycle

4h1991

Lisan al Gaib@scaling01

and normalized by my overall mood

7h627

emi@NoemiTitarenco

@scaling01 this is astrology for NLP people

4h552

Lisan al Gaib@scaling01

@SayItLoud19 agreed

9h1221

Sentio@Sentio_xbt

@scaling01 Anthropic consistently scores higher in sentiment across launches

The crashes during key releases suggest that user expectations are tightly linked to performance promises, and when those aren't met, sentiment drops sharply

12h288

Alex YGift@Radipdegen

@scaling01 scored every single mention but the real metric is how often u typed each one

6.63 for anthropic is generous tbh

12h218

Deepak K@deepakThamizhK

@scaling01 What happened in that Claude 3→4 stretch that tanked your score? gap between 4.8 and 6.63 is wild most people don't quantify their ai bias, you actually did.

10h184