/Tech1h ago

AI commentator @scaling01 argues Claude parameter size estimates of up to 3 trillion are heavily inflated

GPT-OSS-120B shows smaller models can close the proprietary gap.

61110156.8K

#501

Original post

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex#501inTech

@scaling01 true for 5.5 but I am shocked about Opus. They're good this makes me believe in Ant more

Lisan al Gaib@scaling01

Cursor just announced a new 1.5+ trillion parameter model pre-trained on over 100k GPUs.

According to Cursor CEO it is: "as big as Opus and GPT"

Yes, the cat is out of the bag and I can spoil the party. Opus 4.5 to 4.8 and GPT-5 to GPT-5.5 are not that big! (they are all below 2T params)

This means that the current performance of GPT-5.5 and Opus 4.8 is achievable for open-source, because guess what else is that size? -> DeepSeek-V4-Pro

The only moat is scaling. So far, Anthropic is the only lab that successfully made the jump to the ~10T scale, which is also why I no longer see OpenAI catching up until year end. Anthropic can just keep throwing RL compute at Mythos for the next 1-2 years and it will keep improving.

Google went not quite as big and sparsity-maxxed a bit too much. They also have apparently no clue how to RL to make the model actually usable.

OpenAI is still scarred from GPT-4.5

xAI and Meta are still planning.

4:14 PM · Jun 16, 2026 · 2K Views

Sentiment

Users express approval for Anthropic after the Opus size reveal, noting its strong performance relative to larger models like GPT-5.5.

Pos

100.0%

Neg

0.0%

1 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS2.6KBOOKMARKS8REPLIES5

Lisan al Gaib@scaling01

@teortaxesTex GPT-5.5 being larger than Opus shouldn't be a surprise but yeah Ant is pretty good

now I would say: Haiku ~150B Sonnet ~500B Opus ~1.5T Mythos ~6-15T (still massive uncertainty)

Lisan al Gaib@scaling01

my estimate for Anthropic model sizes: - Haiku: 200-500B @ $5 - Sonnet: 700B-1.4T @ $15 - Opus: 1.5-3T @ $25 - Mythos: 6-20T @ $100+

1h2.6K388

LIKES38

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

this is hard to believe, but I've heard similar recently, and in all seriousness… why not? It's not even a question of margins but of opportunity cost in GPU-hours. Mistral-7B > Llama-2-13B, you can just overtrain Moar. if you have the Cards. …Zero upwards scaling progress between GPT-4 and Mythos …Was Elon the only guy who actually bought the scale pill and widely deployed a 3T in 2025? Built a whole datacenter for it, only to become an impromptu cloud provider for Ant's 2x smaller model?

Lisan al Gaib@scaling01

Cursor just announced a new 1.5+ trillion parameter model pre-trained on over 100k GPUs.

According to Cursor CEO it is: "as big as Opus and GPT"

Yes, the cat is out of the bag and I can spoil the party. Opus 4.5 to 4.8 and GPT-5 to GPT-5.5 are not that big! (they are all below 2T params)

This means that the current performance of GPT-5.5 and Opus 4.8 is achievable for open-source, because guess what else is that size? -> DeepSeek-V4-Pro

Google went not quite as big and sparsity-maxxed a bit too much. They also have apparently no clue how to RL to make the model actually usable.

OpenAI is still scarred from GPT-4.5

xAI and Meta are still planning.

1h2K385

Lisan al Gaib@scaling01

@teortaxesTex I think just based on performance, compute and pricing you just tend to overestimate these models

the gap between open and closed models was pretty insane

but GPT-OSS-120B should have been a good hint that they can squeeze much more performance into smaller models

Lisan al Gaib@scaling01

@teortaxesTex GPT-5.5 being larger than Opus shouldn't be a surprise but yeah Ant is pretty good

now I would say: Haiku ~150B Sonnet ~500B Opus ~1.5T Mythos ~6-15T (still massive uncertainty)

1h36570

Lisan al Gaib@scaling01

@teortaxesTex yeah the pricing just throws you off

Lisan al Gaib@scaling01

I had 100-300B for haiku first but increased it because of the price/margin calculation I did afterwards

I think it's MiniMax M2 size ~200B

1h42710