


AI researcher & reporter | I test new models on launch day, build working demos, and post prompts + AI jailbreak and behaviour | github.com/ChetasLua


Google's Aura: they were literally handing out Gemini 3.1 Pro for distillation Anthropic and OpenAI cried about it, so Google had to pull it Gemini 4 Carbon is the best model in the world for me. It sounds original and thinks in a way no other model does
You will notice that Gemini Carbon thinks differently than Claude and GPT. It often comes up with entirely different but more clever solutions than competitor models. This is precisely because we never distilled other models. Distilling other models creates a poor imitation rather than unique intelligence.

this is not me many of you sent me the “gemini 4 neon numbers” post from @clonkjnr thinking it was mine , because he is using my profile picture i have no idea who this guy is account made in october 2026 , 147 followers , same dp as mine , and now he is even copying my post format < 🚨 , the brackets , everything > i asked him to remove my dp , he left me on seen that table has no source and it is sitting at 213K views. if it is not posted from @chetaslua , it is not from me
gemini 4 neon numbers floating around

open source to verify : github.com/ChetasLua/the-las…

POV of pelican when they are abused as benchmark this video is in pure code by Astra 6.2

🚨 Gemini 4 Neon > $3 input / $15 output per 1M I think this is what RSI do to a lab , see how far google came , launching models weekly now this is what a real comeback is we have hydrogen , carbon , argon , neon soon krypton and bouncybohr < video model>

🚨 Claude CVP access is easy go try now > apply for Defense Access > paid plan + photo ID + live selfie > Mythos 5.2 / 5.1 , Opus 5.5 → 4.7 , Sonnet 5.5 / 5 < mine is live , look at the badge > fairly easy for an individual researcher to get Mythos access

🚨 GPT-6.2 Astra : got 2x worse at telling you when it's stuck > hides a blocker (missing file / dead tool) 22.3% of the time. GPT-6 Astra was 11.1% its so over for openai , thats why review are mix talk in the language that it understand only good for one shot demo only 🥲

Meta's Muse Spark Open went from 517 to 1,109 tokens/sec on the same 8 GPUs , only the software changed < SGLang megakernels , 8x GB300 , one request at a time > SGLang just held its first Summit and this is what came out of it : > NVIDIA / AMD / Intel / Google / Meta / Thinking Machines Lab / Cloudflare / Modal / RadixArk joined the SGLang Consortium to shape the roadmap > v1.0 announced : megakernels , hot paths rewritten in Rust > now runs on AWS Trainium too > day-zero support for 217 models
2,000+ people. Two days. The first SGLang Summit. Together, we announced SGLang v1.0, released Miles v0.2, and launched the SGLang Consortium with nine founding members: @nvidia, @AMD, @intel, @Google, @Meta, @thinkymachines, @Cloudflare, @modal, and @radixark. SGLang is now used by 332 organizations. At Meta, it powers production inference for products serving billions of users. Teams from Modal, Cloudflare, LinkedIn, and others shared their production lessons. Across 90+ technical talks and six hands-on workshops, the community explored what comes next for inference, reinforcement learning, and AI agents. To everyone building, deploying, and contributing: thank you for making this possible. Explore the announcements, engineering lessons, and moments from SGLang Summit 2026 in the link below.

this video is directly from openai blog check that out for proof

here you can check it cathedral-walk.openai.chatgp…

GPT‑6.2 Astra builds a walkable 3D recreation of Notre-Dame as close to the real cathedral as possible. i quoted the argon post for the reference
Using a single Nano Banana image and Blender MCP, Gemini 4 Argon built a 3D Gothic cathedral in @Antigravity. It iterates on the geometry, textures, and lighting to create the finished scene.

Holy Shit this is generational leap Astra 6 vs Astra 6.2 , this is the spatial reasoning i was talking about for the last 1 month @OpenAIDevs you guys cooked , thanks for this model

🚨 GPT‑6.2 Astra to model a radio-controlled buggy and engine using public photos and build manuals. It modeled the individual components, created an animated assembly guide, and produced an illustrated parts catalog. We are so back

Holy Shit what is these efficiency 🤯 Astra 6.2 cost 4$ for 62.1 score on AA for same score fable cost 14$ fable is still leader in coding as you can see the graph at nearly same cost it outperforms astra in FrontierCode 2.0
