Pinned Tweet
Break from mind blowing Gemini 4 Argon release I guess lol. Here's the progress on music video generation using Opus 5.5, Midjourney and Elevenlabs. This is V2 with lip sync. More to come plus my thought process, prompt and observations. Needs more work for sure.
44
80
475
47,101
Genie 4: real-time photorealistic worlds you can steer, with physics and native audio.
Trajectory from Pong → this is exactly the kind of progress the simulation hypothesis rests on. Not evidence we’re in one yet — just that the capability side is advancing.
Trajectory from Pong → this is exactly the kind of progress the simulation hypothesis rests on. Not evidence we’re in one yet — just that the capability side is advancing.
The demo above shows how all these capabilities come together in real time. Prompt “traffic light green,” and Gemini reasons through what happens next as the signal changes: the engine revs, and the motorcycle accelerates down the coast.
Full prompt here:
Environment: A sun-drenched coastal highway viewed right at the white stop line of a beachfront intersection, where an overhead traffic light hangs prominently in the immediate foreground glowing bright red. Running parallel to the road strictly along the right-hand side is a golden sandy beach and shoreline with rolling turquoise ocean waves under a clear blue sky.
Character: Shot from a first-person POV seated on a motorcycle positioned right up close to the traffic light, facing straight ahead over the handlebars, side mirrors, and dashboard. With the sea and shore visible on the right side of the frame, the rider rolls the last few feet up to the white crosswalk line and comes to an idling stop directly beneath the glowing red signal, gloved hands resting on the grips while waiting for the light to change.
deepmind.google/blog/genie-4-a…
0
0
1
70
My timeline for next week or so will be all about Kimi K3.1 Pro 3D modeling seems like
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1…
0
0
1
128
This is going to be huge. When Anthropic goes quite for a while and they release something new it usually means it's much more stable with many useful features which would separate them from bot, dot etc. And with Opus 5.5?! Gonna be crazy and cannot wait
We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.
It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.
1
0
2
902
Seems like Codex usage is doing better today? I'm using Astra 6 xhigh for my task and so far doing pretty well on my $200 subscription plan. Did you guys do something recently @thsottiaux @ah20im? 👀
1
0
1
252
Tried to mess around with Astra to make 3D model of new Tesla Roadster mainly this weekend. Still WIP but here's the results after many iterations and refinements
0
0
1
174
DeepSeek V4.1-Flash-1017: stronger benchmarks, faster runs, shorter outputs. 1M context, 384K max output
Pricing (deepseek-flash): $0.15 / $0.60 per 1M tokens (input cache-miss / output) off-peak; $0.30 / $1.20 peak. Cache-hit input as low as $0.003
Big agentic jumps: DeepSWE 77.0, Terminal-Bench 4.0 42.3, ProgramBench 29.1, Automation-Bench 60.5
Optimized for DeepSeek Harness + more
Pricing (deepseek-flash): $0.15 / $0.60 per 1M tokens (input cache-miss / output) off-peak; $0.30 / $1.20 peak. Cache-hit input as low as $0.003
Big agentic jumps: DeepSWE 77.0, Terminal-Bench 4.0 42.3, ProgramBench 29.1, Automation-Bench 60.5
Optimized for DeepSeek Harness + more
🚀 DeepSeek-V4.1-Flash-1017 is here!
Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.
🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, grade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.
1/4
0
0
0
222
Glimpse is Elorian’s visual thinking model. Instead of looking once and guessing, it keeps going back to the actual image/video while it reasons (crops, pulls frames, runs code).
• Matches/beats Astra on robotics, charts, short video captioning
• $0.60 / $3 per million tokens (way cheaper)
• Built by a small team on a few hundred GPUs
Free to try + API: platform.elorian.ai
x.com/AndrewDai/stat…
• Matches/beats Astra on robotics, charts, short video captioning
• $0.60 / $3 per million tokens (way cheaper)
• Built by a small team on a few hundred GPUs
Free to try + API: platform.elorian.ai
x.com/AndrewDai/stat…
Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that.
Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.
0
0
1
116
Roadster, as promised the most insane production car of all time! 🤯
Roadster transforms in Track Mode
These aero features aren't just for looks: they provide the best passive downforce package of any road car (approaching F1 levels!) with wing actuation driven by the @Tesla_Optimus actuator system
0
0
6
2,085
Probably won't happen this week but most likely next week! This can really nuke OpenAI though and as I said many times they have to act fast and give us better models and not just bunch of features. Features are good but definitely don't attract people as much as beast models!
Anthropic have an incredible opportunity to absolutely Nuke OpenAI
Drop Fable 5.5 this week and make it fully available on Pro plans, cut the pricing to beat Astra + it being more token efficient
OpenAI are still scrambling to fix their safety issues that caused 6.1 Astra to get shelved so are out of the game for a short while
0
0
1
1,256
Guess it's time to test it out and compare!
For Those Who Create. KLING 4.0 Is Now Fully Available. 🎬
What’s New, Starting Today
🖥️ Access Kling 4.0 through a refreshed creation interface on kling.ai
🔗 Access Kling 4.0 through MCP and CLI for greater flexibility in your video creation workflow
Coming output at 4K and 1080p resolutions
🛠️ API access for enterprise customers and developers
You call the shots. Let the creating begin!#Kling4 #KlingAI
0
0
0
101