


A smol bee 🐝 :)



Okay this one is pretty interesting.. FrontierRSI is basically testing if AI can do the actual work of a researcher at an AI lab.. pre-train, post-train, inference, harness and even kernels! Most benches just give it basic ML tasks so this is a lot closer to the real thing Can't wait to see where Gemini 4 Krypton and Fable 5.2 land on this!!
Introducing FrontierRSI, a new recursive self-improvement benchmark by @abundant_labs. We evaluate agents on (1) a broad range of AI research and engineering work, and (2) post-training models on multi-step agentic tasks, rather than single-step tasks as in PostTrainBench.

Gemini 4 Neon is here.. Oh no it's Genie 4.. Don't worry guys Gemini 4 Neon is still coming!
Genie 4 is now available in Project Genie. It brings native audio, real-time editing, and a new agent to help you create interactive, virtual worlds.

So it's Gemini 4 Neon day then.. Model has been spotted on Google Cloud's billing catalog and in Antigravity CLI ~ pricing • $3/M input and $15/M output • Batch and Flex at half, $1.50 in / $7.50 out • Priority at $5.40 in / $27 out • input takes text, image, video and audio ~ Antigravity CLI 1.5.11 • new "gemini-pro-orchestrator-neon" model, a Gemini Pro orchestrator Not answering anywhere yet.. It is a weird model tbh it sits between Carbon/Argon < Neon > Oxygen What is google doingg.. ima have to learn the periodic table atp..
Gemini 4 Neon has been added to Antigravity CLI as an Pro orchestrator model.

I have some bad news for Free and Plus Gemini users.. It seems like you won't be getting Gemini 4 Hydrogen in Antigravity ~ before • Hydrogen was on Free, Plus and every Pro plan, trials included ~ now • only on paid Pro, Ultra and Enterprise • Pro trials lost it too • Claude 4.6 and GPT-OSS 120B also get removed today, so Free and Plus are left with just Gemini 3.8 Flash ~ Google One support confirmed it to a friend of mine • JIO and other promo AI Pro trials don't get Hydrogen • only the trial that comes with a premium device still does • they offered 1000 AI credits instead - All this conclusion is from what I've seen on the official docs and the support mail and I'd love if some clarification can be provided on this.. @_mohansolo

It is the best time for Google to drop Gemini 4 Krypton! OpenAIs launch was pretty underwhelming.. it got a lot of negatives.. and it's the perfect time to launch a banger and just put it all away to dust.. Ofc it could be Flash-lite aka Neon or Nano Banana Pro 2.. I'm not entirely sure but so much hype indicates something big so let's see..
🚨Breakingg! Logan has spoken Gemini!! What's coming todayy?? We have 11-12 hrs before launch..

🚨Breakingg! Logan has spoken Gemini!! What's coming todayy?? We have 11-12 hrs before launch..
Gemini

Antigravityy Resetss!! Today marks 1 year since Antigravity was launched and to celebrate that the team has reset the quota for everyone!
One more thing… In honor of Antigravity turning 1, we’re resetting quota for everyone across all models!

Wow!! Antigravity team is on a cooking spree too Since carbon was launched the (TTFT) was pretty high.. in simple words it's how long you wait after hitting send before the model starts writing its reply.. Remember lower is better! And now it's 1.5x faster on Antigravity for Carbon and will be faster for Argon too soon.. This is the kind of improvements we all wanted from Google & Agy and I'm glad that the team is working on making the whole experience better too!
Time to first token (TTFT) for Carbon is now 1.5x faster on Antigravity. Will follow soon with Argon too. Lots of work went into this so you all could have a faster experience. More to come 👀


Thanks to @long_longVN for testing out my prompt! and I'm waiting on @notjazii to get me a sub.. come on brooo :)

My first test with GPT-6.2 Astra and Sol - Mechanical Bee Astra is clearly far better than Sol here and that's expected.. ~ Astra took 1h 12mins and Sol took 43mins ~ Astra used 19.8M tokens ($14.91) and Sol used 14.8M ($3.84) Astra did so well in polishing the bee model and all the placement of parts <things just make more sense than sol> The overall UI design is covered by my prompt but both models did a terrible job <too much space and bland design> Astra still has that greenish look in the dark theme even tho it was not mentioned to use any kind of green tint.. Overall both are great at 3D.. definitely cheaper than Fable 5.2 but design taste is just not there yet.. My recommendation, you'll be better off using Sol 6.2 for 80% of tasks as with enough steering it can achieve significant or better results than Astra while still being cheaper than it and quicker!

Another one lol.. this is crazy
I will not be writing a full review of GPT-6.2 Astra / Sol. I had a terrible experience testing them. They were essentially unusable. I do not recommend these models.

Omg this is not looking good.. Theo hates GPT-6.2 Astra.. I get that he's a bit extreme in his takes and I do not like all of them but his reviews are not bad at all and you literally can't review models better than him.. ~ the problems • on long runs it blocks itself to ask you internal stuff you can't possibly answer, like which clip file a subagent wrote to • explanations are so deep in the weeds he switches to Opus 5.5 just to get a readable summary • runs in loops, 7 days on his ts-rust project with 0 improvements and it made the compiler 56% slower ~ what he still likes • great reviewer, catches stuff other models miss • world class at 3d and game work • computer use is nuts His prediction is people love it day 1, turn on it day 2 and forget it by day 14.. I'm doing a test with Astra 6.2 let's see how it does on it..
Theo - t3.gg @theo Nov 17 Why I hate GPT-6.2 Astra During early testing, I found myself so frustrated with this model that I spent a day writing a letter to OpenAI about it. They fixed a small handful of my complaints, but not enough for me to

NVIDIA, AMD, Google, Meta and Intel all backing the same open source project is not something you see every day.. They just joined the new SGLang Consortium along with Thinking Machines Lab, Cloudflare, Modal and RadixArk to help shape its roadmap. SGLang v1.0 was announced too, built for running models in production at scale. And it now runs on AWS Trainium in public preview..
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.