I told Claude Fable 5.2 and Opus 5.5 to build a Porsche 911 GT3 RS cinematic in Blender.
In a dark showcase studio, the car assembled itself piece by piece across 11 exterior and interior camera shots, ending in a 34-second render.
Everything came from prompts and reference photos; I only directed fixes along the way. I never touched Blender.
👀 Gemini 4 Neon just got its official listing on Google Cloud.
Google calls it a fast reasoning model. It's built on Gemini 4 Hydrogen with deeper reasoning, and they say it gets close to Carbon on coding, reasoning and agentic tasks at Flash-tier speed and cost.
GPT-6.2 Astra just landed #3 on Code Arena WebDev and Sol #7, both on the pareto frontier
Astra is +18 over GPT-6 Astra and #1 in Consumer Product while costing $5.45 per task instead of $9.46.
Sol sits 24 pts behind at $1.45 a task ngl Sol might end up being the one most people actually use x.com/arena/status/212311269…
Genie 4 is out.
You can build a world, hear it, edit it while you're exploring it, and have an agent help you shape it. Then anyone can take your world and remix it into something new.
Game prototypes, interactive stories, little worlds just for fun. People are going to make amazing things with this.
Only on AI Ultra for now, but I really hope it reaches more people soon.
Anthropic might already be testing the next Fable model.
A stealth model called Silicon 2 reportedly appeared inside Devin Build, sitting in the Claude model lineup.
> 4 reasoning tiers: Low, Medium, High, XHigh
> $10 input / $50 output per 1M tokens
> Fable 5.2 is $8 / $40 for comparison
Fable 5.2 only dropped on Nov 9, so this could also be something else entirely. Still a rumor until Anthropic says a word.
But with GPT 6.2 just out and Gemini 4 Neon close, the timing makes a lot of sense.
OpenAI just launched ChatGPT for Sales
It connects to Salesforce, HubSpot and Zoho so it can work directly with your pipeline, accounts and deals
ChatGPT now has versions for each department
gpt 6.2 sol (colosseum test)
very impressed with this one ngl, took 15 minutes to get this done
this time it didn’t get it right in one shot unlike my last test on the 3d chopper.
Gemini 4 Neon appeared in Google Cloud’s SKU catalog.
It’s priced at $3 per million input tokens and $15 output. That’s exactly twice Hydrogen. Batch and flex cut it in half, priority lane goes higher.
No official announcement yet but same pattern we saw with the previous Gemini 4 models before they launched.
Another one-word "Gemini" tease from Logan. Gemini 4 Neon just got pulled from the live Gemini web config while Argon and Carbon stayed.
Neon, Helix and Krypton are all rumored for the next drop.
Nothing official yet. But Google is shipping a new Gemini 4 model every few weeks now. Astra barely got one day in the spotlight.
🚨OMMGGG!
Microsoft dropped 7 new MAI models at Ignite, and MAI Code 2 is already sitting in GitHub Copilot.
> 96% of Opus 5.5's coding performance at 5% of the price
> 98% of GPT 6.1 Sol at 40% of the price
> Preferred over GPT 6 Sol 64% of the time in blind frontend tests
Microsoft is upfront that it isn't built for long agentic work or hard science yet. But for everyday fixes and features, running agents all day without burning money finally looks possible.
microsoft.ai/news/microsoft-…
My honest take on GPT 6.2 after a full night of testing.
The intelligence is clearly there. Sol built me a fully flying 3D helicopter with assembly and exploded views in 7 minutes, and it stayed fast the whole time. The design taste is a real step up from 6.1 Sol.
But once I moved away from 3D, things changed. I had to keep rewriting prompts, add more context and push it to continue before I got results I was happy with. A model this smart shouldn't need that much effort from the user.
I think that's why the timeline is so divided. Most early tests are 3D, where it shines. Throw everyday work at it and the experience is very different.
I'm not the only one seeing this. @sama said they still have work to do on making the models nice to work with, and that it's the priority for upcoming releases. @TokenGremlin pointed out that Astra takes shortcuts whenever a prompt leaves room, like grabbing a ready-made asset instead of building it.
The good news is @thsottiaux says the team never stops working on it, and it's still a solid jump on every front while being cheaper and faster.
so i’m hopeful. the intelligence is already there, they just need to make it easier to actually use.
if OpenAI gets that part right, millions of people might come back.
starting to feel like openai just juiced the 3D on 6.2
like 80% of early model tests are 3d stuff so of course everyone thinks it's cracked
then you try it on anything else and it's slow af lol
GPT 6.2 Sol built this helicopter in 7 minutes 🤯
A fully flying 3D chopper with assembly and exploded views, and the design taste is on another level compared to GPT 6.1 Sol
Said it two days ago. We got the nerfed GPT 6.2 after all.
I tested 6.2 Sol on medium 3D designs first and honestly it was great. Good output, good design taste, no complaints there.
Then I tried it on other stuff and it was painfully slow. Simple tasks took way longer than they should, and that's hard to work with every day.
@TokenGremlin called this a few days ago. He said 6.2 would show up loaded with guardrails and nerfed from day one, and that's pretty much how it feels right now.
OpenAI clearly rushed this one out. Hope they do something about this. Big disappointment for real.