RedMoon

Zero brand loyalty. Building in AI, testing everything, roasting the bloat. Real production workflows over marketing demos.

[RedMoon @RedMoon1484672]

https://t.co/u1ZIGgferr

[RedMoon @RedMoon1484672]

Europe’s "sovereign AI" champion just dropped Mistral Large 4 to reclaim Western independence.
In their official coding benchmark, every single rival is Chinese—and Kimi K3 still beats them 68% to 62%.

[RedMoon @RedMoon1484672]

https://t.co/6GlfBShG1l

[RedMoon @RedMoon1484672]

@Taniyatweets_ You don't need to.

[RedMoon @RedMoon1484672]

@IndependentEco Another possibility is that Anthropic already has a product that's even better than Fable 5.5

[RedMoon @RedMoon1484672]

@haider1 I think AI is still in its infancy right now, but in another five years, its development will definitely exceed our imagination - maybe it'll even start to take the initiative on its own

[RedMoon @RedMoon1484672]

@buildwithrajath OpenAI's strategy has huge problems. They probably feel like they're losing the competition with Claude in terms of model capabilities, so they've started to shift their focus elsewhere

[RedMoon @RedMoon1484672]

@JamesMalsawm And that’s exactly why everyone is ditching OpenAI.

[RedMoon @RedMoon1484672]

@Seltaa_ I agree, chatting and work shouldn't be mixed up

[RedMoon @RedMoon1484672]

@mjjwiki 甲骨文云比较简单

[RedMoon @RedMoon1484672]

AI chips get so hot that just simulating the heat is now a $1.5B business.
Vinci raised $250M to take on Synopsys and Cadence, promising physics models that run 1,000x faster.

[RedMoon @RedMoon1484672]

@heyclairelin So I'm here to support you.

[RedMoon @RedMoon1484672]

Is this the ultimate case of having your cake and eating it too?
Within 24 hours, Honda and Nissan both tapped Applied Intuition for vehicle OS while refusing to give up architectural control.

[RedMoon @RedMoon1484672]

@mikeydsoftware I think it makes no sense. You can't rely on marketing to make up for it.

[RedMoon @RedMoon1484672]

@mikeydsoftware Neither. Expect strong benchmarks, but tool calling decides if it's usable.

[RedMoon @RedMoon1484672]

The most reliable trade in AI is selling industrial air conditioning.
LG just locked in a 5GW deal for North American data centers, pushing its order backlog to 3 trillion won.
In comparison, Claude and OpenAI are still losing money.

[RedMoon @RedMoon1484672]

OpenAI is panicking. Tibo promised Codex will either drop a new feature or reset rate limits daily.

Optics-wise, OpenAI is losing the foundation model race. Astra launched with massive hype, only for Opus 5.5 to completely blindside them.

Opus 5.5 is incredible at animations and coding. I always thought 3Blue1Brown-style videos had a high barrier because video models lack mathematical precision, and LLMs lack camera intuition.

Anthropic actually pulled it off. It feels like an LLM learning to drive itself. There must be massive alignment work behind this. Anthropic is massive yet nimble, always ready to drop something huge.

OpenAI needs real breakthroughs on the model side. Shipping minor Codex features is cheap. Codex ships one a day, but anyone can build ten wrappers a day.

Model and shell must align. When you fall behind on the model, fix the model. OpenAI is using centralized UI wrappers to patch local workflow problems.

OpenAI models are already slipping into tier two, and they have lost the confidence to even hype the next one.

[RedMoon @RedMoon1484672]

@giadotai I definitely don't trust Claude's ethics either, but right now there's simply no model on a higher level than theirs.

[RedMoon @RedMoon1484672]

@JamesMalsawm Opus 5.5 is sitting at 58. When Fable 5.5 drops, Anthropic sweeps the top tier.

[RedMoon @RedMoon1484672]

Subscribed to the $500/mo ChatGPT Pro to test it out.
If what Tibo said about subscription and API pricing converging in the future is true, I might downgrade and just go with the API.
First impression: quota burns crazy fast. Ran 40 minutes of standard-speed, single-threaded GPT-6-Astra codex, and 4% of my quota is already gone. That's 1% every 10 mins on average.
Burn rate honestly feels about the same as the old $200/mo tier. Massive price hike. But I noticed that without upgrading, calling GPT-6-Astra on the old $200 plan constantly threw the error "Selected model is at capacity. Please try a different model." At least upgrading reduced it—just reduced, not completely gone. Ran it for a bit and it started popping up again. Fuck.
Goes to show OpenAI is definitely bleeding cash + compute-starved lately.
The only right way to use ChatGPT Pro moving forward: offload whatever you can to Dot. Only burn quota locally when you absolutely have to create Codex tasks locally. Otherwise, even a $500 sub won't be enough.
OAI is an absolute ripoff.

[RedMoon @RedMoon1484672]

Lmao, Google finally folded.
Starting Oct 9, personal accounts without a Google AI plan are stuck with just Flash-Lite—both Flash and Pro are getting locked behind a paywall.
Looking at what other AI labs have been pulling lately, tokens aren't getting cheaper; they're getting pricier.
Everyone's quietly finding ways to nerf subscription limits.