A post appeared on my feed with 2.8K likes. It said a model called Claude Next is being tested inside Claude Code right now. It makes images, video, and music that don't look generated.
Pure JavaScript, the leaker claimed. "It is over for image and video generation."
I checked the account.
No Anthropic confirmation. No release notes. No blog post.
Just one tester on X with a history of spotting unreleased models, plus a video of something animated.
My first thought: here we go again.
Anthropic has not said a single word about Claude Next.
Not confirmed. Not denied. Not even a wink.
The company just shipped Opus 5.5 and Sonnet 5.5, and the internet has already moved on to a model that may not exist.
That's the part that gets me. Not the leak. The speed of the leap.
A rumor becomes a headline becomes a product in people's minds in about eleven hours.
Why should you care? Because this is how the AI news cycle works now.
It doesn't start with a launch. It starts with a screenshot and a confident tweet.
And everyone, including people who should know better, starts planning their lives around it.
I spent the evening reading the thread.
The reactions. The skeptics. The memes.
Here's what i found.
Two models, one name
here's a question people always ask: what even is Claude Next?
Trick question. It's two different things wearing the same name.
Three years ago, Anthropic's IPO deck mentioned a frontier model, tentatively called Claude-Next. Planned to be ten times more capable than the best model of the day. Training cost: around one billion dollars, spread over an 18-month cycle.
That was the old Claude Next. A budget line. A promise.
The new Claude Next is a leak. A tester on X says it shows up inside Claude Code when you're not expecting it. It supposedly burns tens of millions of tokens on a single generation.
It makes Opus 5.5 look primitive, in his words.
Same name. Totally different thing.
One is a plan. The other is a rumor wearing a plan's clothes.
That confusion matters. When people argue about whether Claude Next is real, they're not even arguing about the same thing.
| Old Claude-Next | Leaked Claude Next | |
|---|---|---|
| What it is | IPO deck plan | X posts from a tester |
| Capability claim | ten times today's models | makes Opus 5.5 look primitive |
| Evidence | investor materials | screenshots and a video |
The leak treadmill
what actually happens is a pattern i've watched repeat for a year.
- A codename shows up in a model selector or a tool like Cursor.
- A power user screenshots it.
- The AI community goes feral for 48 hours.
- The model ships under a boring name, or everyone forgets.
The list from the last year alone:
- Honeycomb EAP
- Mythos
- Fable 5.5
- wafer EAP
- and now, Claude Next
Some were real. Some were A/B tests. One of them, people joked, was probably just Opus with a new temperature setting.
The best reaction I found to the whole circus was a parody leak. Someone announced Claude opus-Optimus-prime-9000, coming next Friday, with the caveat that Gemini told them. It got upvotes.
That's where we are. The community has leak fatigue, so it started leaking fake leaks to cope.
But here's the thing about Claude Next specifically. The timing is odd.
Video without a video model
i used to think generated video meant a video model. Diffusion, frames, the whole pipeline. Claude Next, per the leak, does something else.
The claim: it generates animations, visual scenes, and physics-based motion purely through code.
JavaScript. Canvas. Web tech.
No native image model. No native video model. The media comes out of the model writing code that renders it.
If that's true, it's a strange and interesting bet. OpenAI and Google both ride dedicated models for images and video. Anthropic's rumored answer is: make the language model so good at writing rendering code that you don't need them.
I cannot verify any of this. Neither can you. Zero official confirmation.
The most honest summary I saw on Reddit was a comment with 18 upvotes:
i don't know what bro was saying in that tweet, but cool.
Still, I get the appeal. Code is compact. Code is inspectable.
A model that renders through code can be fixed and reused in ways a generated mp4 can't.
Claude is building Claude
here's the detail that actually made me stop scrolling. It wasn't the leak.
Anthropic says its own models now lead 26 percent of the company's AI research and development. Over 90 percent of the work involves Claude in some capacity. Claude helps build the next version of itself.
So Claude Next is a leaked name and a description of the workflow at the same time. Whatever it ends up being, a meaningful chunk of it will have been made by the thing it replaces.
Nobody knows what that does to a model's design. Not the public. Not really the company either, i suspect.
Everyone is making this up as they go, including Anthropic.
And that, more than any anime video, is what i find worth watching.
About the codenames
Tangent, but stay with me.
Everyone names AI models like fantasy characters now.
Mythos. Honeycomb.
Fable. Next.
My home directory is a museum of names i gave up on: fix_shit_v2.py, final_final_2.py, script_OLD_DO_NOT_USE.py.
I think about this every time a codename leaks. The people building these models get cool names because the models ship and get famous. The rest of us name our work the way you name a fish you're about to eat.
There's status in it. Leak a model called Mythos and you sound like you're inside the story. Tell me about your script called temp3 and I know exactly what your life is like.
Both are honest. Only one gets the likes.
Anyway. Back to the model that may or may not exist.
Who should ignore this
Let me be blunt. Most people reading this should not care about Claude Next yet.
It is not released. It is not confirmed. No access, no waitlist, no price, no API.
If you're not a power user living inside Claude Code, this leak changes nothing you can do tomorrow.
The biggest misconception is that Claude Next is a thing. It is a codename that has floated around for three years, plus one very good tweet. That's the entire public evidence.
The hype machine runs anyway. People are literally betting on which model ships next. Prediction markets track release dates like they're game scores.
That pressures a company to ship before a model is ready, and we've seen how that ends more than once.
Do not buy a subscription because of a rumor. Do not cancel one either. The announcement is free, and it lands eventually.
i still don't know if Claude Next is real. Maybe it ships next month as Opus 5.6. Maybe the leak was a clever test.
Maybe Anthropic is quietly letting the community do its marketing for free.
If it ships and the code-rendering stuff holds, I'll be genuinely curious. I'll even admit I was one of the people refreshing feeds all evening.
But the most honest sentence I found all day was the top comment on that Reddit thread: i don't know what bro was saying, but it looked cool.
That is the whole AI news cycle in one line.