NZ ☄️ retweeted
🚨 A "gemini-3.8-flash-router" slug has reportedly been sitting in Antigravity's metadata for a couple of days.
I'm increasingly convinced the next Flash ships as a router, sending easy prompts down a cheap path and hard ones to a heavier model.
Maybe it's just an internal test name, but right now that's the cheapest way for Google to look fast and smart at the same time.
Gemini 4 Flash might be in grey testing for some users..
And its pretty fast while being extremely close in capabilities with Gemini 4 Argon.. could be even better when its released..
Carbon is looking pretty good too from what we're seeing on arena..
I'm not getting any more outputs of it since yesterday.. so a new checkpoint might be coming soon..
NZ ☄️ retweeted
Today, I'm launching llmdojo.dev:
- The hands-on way to learn AI engineering: LLMs, RAG, agents, evals and inference.
- 50+ interactive labs. Tweak tokenizers, attention, embeddings and KV caches live.
- Train a neural network in your browser and watch it learn.
- Build agentic workflows with tool calls, memory, guardrails and bounded autonomy.
- Ship RAG that works: chunking, hybrid search, reranking, and LLM-as-a-judge evals.
- Production AI: model routing, fallbacks, quantization, caching and cost per request.
- 490+ interview questions, reported from OpenAI, Anthropic, Google, Meta, Amazon, Microsoft, Mistral and 50+ more.
- Every answer is cited from papers and books, or measured on real models.
NZ ☄️ retweeted
is there a secret startup everyone's joining or what?
anthropic and openai people leaving back to back this week
NZ ☄️ retweeted
I WAS SO HYPED FOR FABLE 5.5 NGL
but people testing the checkpoint say it's ONLY A BIT BETTER than opus 5.5.
meanwhile kibitz (kimi k3.1) and kalmia (
Qwen 4) are cooking on design for WAY LESS MONEY
if fable comes in at a CRAZY PRICE FOR A SMALL JUMP, a lot of us are just gonna stay on opus or go cheap
NZ ☄️ retweeted
security teams right now wishing cve.now dropped yesterday
NZ ☄️ retweeted
Claude reset is here, LFG! 🔥
...and i'm already at 14% on a $200 plan. how?? 💀
Anthropic, these session limits make it hard to do real work. wadahelly
I do have Codex, but i barely use it. Last time it felt kinda dumb. Opus 5.5 is just on another level.
NZ ☄️ retweeted
late to this but WOW
just woke up and anthropic dropped another product. 😭
claude duet works for you even when you step away, and still asks before it does anything big.
openai dots and grok bot got real competition now
NZ ☄️ retweeted
I have a theory.
What if Anthropic is deliberately holding back Fable 5.5 because they don’t want OpenAI to launch GPT-6.1 Astra right after them?
NZ ☄️ retweeted
This is why DeepSeek gets so much love.
They share their research. They share their models. Now they're sharing the engine that runs them in production.
I don't see the big US labs sharing their production setup like this.
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
NZ ☄️ retweeted
2.8x over stock trl, and to their credit the like-for-like bar is right there in the chart
stock trl zero-3: 3,885 tok/s/gpu
halo ep1 zero-3: 5,560
halo ep2 zero-2: 10,479
~1.4x at matched sharding, ~2.7x if you move to zero-2 as well. both real, different claims
just starred the repo aswell👇
Introducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.
Star us on GitHub: github.com/whitecircle/halo
NZ ☄️ retweeted
one 3d racing game. 125M tokens.
Cost: $0
built by space bunny, the stealth model nobody has claimed yet
coding: four seasons, traffic, crashes, gears, mobile controls. first playable build in 10 minutes
multimodal: don't explain a feature, sketch it. send the photo and it builds from that
fast: 90 tok/s early on
1M context: load a whole repo in one go
same tokens on claude opus 5.5: ~ $512 (~$47 with caching)
here's the game:
NZ ☄️ retweeted
🚨 Anthropic's next big model might be closer than we think.
Report says Anthropic is about to release a new model with sharper reasoning and intuition, and it can tell what you actually want even when you barely explain it.
My bet is Fable 5.5, and since Fable 5.1 also dropped on a Tuesday, tomorrow wouldn't surprise me.
But honestly, the only reason I see to launch it now is the upcoming IPO. Opus 5.5 is already the best model out there with no real competitor. They should hold Fable 5.5 until a serious rival shows up, like Astra 6.1.
And it may not come alone:
> Fable 5.5
> GPT-6.1 Terra
> Gemini 4 Carbon
Add the next wave of Chinese models and this could be the craziest week in AI yet.
