noclipepe

ff

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

Your AI startup could be obsolete before it ships.

Sam Altman says the next 6 months may deliver as much model progress as the last 2 years.

If he’s right, today’s impossible feature becomes tomorrow’s API—and today’s moat becomes a model update.

Don’t build for the frontier you see.

Build for the one arriving next.

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

Sam Altman never says “open weights” in this clip.

But he frames the real argument behind them better than most.

AI can become the greatest distribution of power we’ve ever seen—or give one company, person, or model more power than everything else on Earth combined.

Listen to the final 10 seconds.

The question isn’t which model wins.

It’s who gets to decide.

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

Your AI coding agent isn’t bad.

You just sent it to work with its hands tied.

Same prompt. Same game-cloning task. Two agents.

No MCPs: a functional but basic-looking prototype.

2 MCPs: a far more polished result.

The second agent generated images with Higgsfield, passed them to Meshy AI, then used the assets in its build.

MCP didn’t make the model smarter.

It gave the agent an art pipeline.

Watch the comparison.

Are we still benchmarking models, or the production teams connected to them?

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

@unchosen_one fair

[noclipepe @noclipepe]

Why does each new Opus version cost more if Anthropic has kept its pricing flat?

Same two Three.js tasks:

Opus 4.7: 295k tokens, $7.41
Opus 4.8: 392k tokens, $9.80
Opus 5: 407k tokens, $10.19

The token price never moved.

Token consumption did.

Opus 5 also took 93 minutes, versus 35 for 4.7 and 31 for 4.8.

Then the video flips the verdict.

In this run, Opus 5 was the only one judged to match both landmark references and dimensional requirements.

Would you pay more and wait longer for the result you would actually ship?

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

@LmbtnE same lol

[noclipepe @noclipepe]

OpenAI taught speech-to-text what to expect before anyone speaks.

Its new GPT Transcribe can take recording context, names, jargon, expected languages, and previous Realtime turns before it listens.

OpenAI says it cut Whisper’s error rate by 52% on a 22-language benchmark.

GPT Transcribe costs about $0.27 per hour. GPT Live Transcribe handles low-latency audio at about $1.02 per hour.

The benchmark still needs independent testing.

The next voice AI advantage may be designed before the first word is spoken

[OpenAI Developers @OpenAIDevs]

https://t.co/OBMh53HIkU

[noclipepe @noclipepe]

GPT used the fewest tokens. Grok made the biggest spectacle. Kimi beat both.

Same task. Same React stack. Fresh terminals.

GPT-5.6: 19.2K tokens, $0.30. In the canyon scene, the cars landed intact.

Grok 4.5: 55.2K tokens, $0.45. Flying wheels, folding bumpers, pure Hollywood.

Kimi K3: 32.7K tokens, $0 on the API meter because it ran locally on 8× B300. The wrecks stayed deformed, and the cars fell into the gorge.

Kimi won all 3 scenes for me.

Watch the video and tell me I’m wrong.

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

@Kamikaze6888 yeah definitely seeing a pattern here

[noclipepe @noclipepe]

In this round, Opus 5 completely destroyed Fable 5.

Same prompt. Same test: "Build an ultra-realistic 3D flight simulator."

The gap wasn't even close:

Fable 5 ($2.82): 2x more expensive — broken aerodynamics, laggy controls, 2008-era graphics.

Opus 5 ($1.40): 2x cheaper — realistic bank/pitch physics, smooth WebGL lighting, instant response.

Opus 5 just completely reset the baseline for AI-generated 3D graphics.

Which flight physics look better to you?

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

@baderslashxqt fair enough

[noclipepe @noclipepe]

Claude Opus 5 just got crushed on speed by a model that cost under $1.

Same prompt. Same React stack. Fresh terminals.

Fable 5: $0.98, 310 seconds
Kimi K3: $0.78, 1,512 seconds
Opus 5: $2.21, 1,529 seconds

Fable 5 was 5x faster than Opus 5 and Kimi K3.

Kimi K3 wins on raw API cost ($0.78).

But speed and price are not what you ship.

Watch the results. Which model actually won?

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

Four AI configuration. An insane 3x gap

The build times were nowhere close:

Opus Max: 10:36
Kimi K3 Max + Pi: 16:56
Kimi K3 Max: 17:44
Opus Medium: 32:48

The timer has an obvious winner.

The video might give you a different one.

Which result would you actually ship?

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

Anthropic is still outside the Open Secure AI Alliance.

OpenAI joined.

Google joined.

They joined NVIDIA and 36 founding partners in the Open Secure AI Alliance, launched today to share AI security models, agent harnesses, tools and research openly.

The bet is simple: defenders should be able to inspect, adapt and run frontier systems locally, without one API provider setting the limits.

The divide is now impossible to ignore:

Open defense or gated capability.

Does the last missing frontier lab join next?

[NVIDIA @nvidia]

AI security advances when the industry builds in the open, together.

We're introducing the Open Secure AI Alliance with industry leaders to develop new techniques and tools to safeguard software and agents.

By sharing models, tooling and research in the open, we can broaden the community of defenders.

Learn more about the founding members’ contributions: https://t.co/A16oqxs5Ty

[noclipepe @noclipepe]

The open-weight era just crossed a new line.

Moonshot published the document explaining what is actually inside Kimi K3.

2.8T total parameters.
104B active per token.
16 of 896 experts selected.
Native vision.
1M-token context.

A model at this scale can now be inspected, modified and built on.

Open weights are no longer the side story.

They are becoming the main event.

Read the report before your feed turns it into recycled takes

[Kimi.ai @Kimi_Moonshot]

Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.

Model weights: https://t.co/7m7eEg6Y0B
Tech report: https://t.co/yeu6cjpMCT
Tech blog: https://t.co/YTfiMSNM1f

[noclipepe @noclipepe]

@knowix_eth And it was the cheapest one tho. kinda wild

[noclipepe @noclipepe]

@Saidi_Token that’s the leaning tree of Pisa actually

[noclipepe @noclipepe]

This video makes the AI leaderboard look useless.

Three frontier models got the same prompt. Each won a different game.

Opus 5 built the richest world: varied terrain, wind-reactive banners, unit animations and hit reactions.

GPT-5.6 Sol had the best UI, according to the author.

Grok 4.5 finished fastest by a wide margin in this run.

Same React stack. Fresh terminals. No external visual assets.

There is no best model.

Which result would you actually ship?

[noclipepe @noclipepe]

https://t.co/svltbukskC

[noclipepe @noclipepe]

@Im_IrushiK Fair. But Fable was 5x faster. Doesnt that speed gap count for sth?

[noclipepe @noclipepe]

@Kestrelgdy Facts. One wins your time, the other wins your budget

[noclipepe @noclipepe]

@vin_diesel__fan More comparisons? Ofc
I dont turn my back on family