π¨βπ»Engineering AI systems for AJ Hackett Bungy in New Zealand ποΈ Building dev tools in public.
New Zealand
Joined June 2022
- Tweets3.6K
- Following994
- Followers1K
- Likes3.2K
Only google would announce a model and not release it to everyone.
I did say we'd be releasing our models periodically...introducing Gemini 4 Carbon, our next frontier model.
Following Argon's release just two weeks ago, Carbon shows the pace at which weβre pushing the frontier, it ranks #1 on the recently updated v5.0 of @ArtificialAnlys Intelligence Index at 59. It also leads across key cybersecurity evaluations scoring 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
Weβre seeing significant improvements in long-horizon tasks even compared to Argon, and weβve strengthened frontier safeguards.
As we did with Argon, weβre going to roll it out gradually + responsibly. Some trusted cyber defenders will get access starting this week. We'll be expanding it to more partners through our Fairwind Program, ahead of wider availability soon, so hang tight!
which open model are you actually waiting on
glm 5.5 flash
qwen 4 27b
kimi k4
something else (say below)
mine: Fable 5.5!
Matt retweeted
Introducing π»π΅πΌπΉ: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> π»π΅πΌπΉ-π·.π½π±: text, 256k context window.
> π»π΅πΌπΉ-π
π»-πΈπ±: vision-language model with video-input support for spatial understanding
> π»π΅πΌπΉ-π°ππππ-πΈπ±: tool calls, voice chat, reasoning in 18 languages.
> π»π΅πΌπΉ-πΎπππ-πΈπ±: an omni model combines vision, audio, and language.
> ππ·-πΈπ± and ππ·-πΎπππ-πΈπ±: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> π»π΅πΌπΉ-π
π»-πΈπ± is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> π»π΅πΌπΉ-π·.π½π± is #1 on the Pareto frontier of device intelligence πππππππ.ππππππ.ππ
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ π§΅
a lead qualification bot doesn't need frontier reasoning
it needs to answer fast and not sound like a robot
small open model tuned on your own calls > big model with an 800ms pause
@hopper_research are betting on exactly that, gemma 4 31b at 80ms ttft πa
Thereβs a huge blindspot in the inference market today: itβs built almost entirely for coding agents.
Voice is how billions of people will interface with AGI. Thatβs what weβre building at @hopper_research : the inference cloud for voice
Excited to launch today at withhopper.com
who wants to be a millionaire?
i actually want to know if your account on X is worth $1M
go check this out myaccworth.com
i'm in christchurch so my odds of bumping into a vc over coffee are basically zero
but if you're raising in sf, someone just mapped every cafe they hang out at
check it out! β
meetvc.cafe
made a map of Bay Area cafes where youβve got the best shot at meeting VCs (and maybe raising your next round)
check it out: meetvc.cafe
how much do you spend on AI coding a month?
$0: free models + opencode
$20: one sub
$100-200: max plan
$500+: you're either shipping or cooked
mine: $30 π
Everyone's setting up Grok Bot. Then they look at the empty box and think: "ok... now what?"
So I made 5 ready-to-use bots you can copy and paste:
π₯ Inbox brief
π
Meeting prep
π§Ύ Weekly expense sweep
π Price/listing watcher
π Research-to-doc
Free: codermatt.gumroad.com/l/vaoeβ¦
How to connect @Instaboxsh and @AmpCode using self-hosted orbs!
Matt retweeted
$100k to the developer who can make muse spark open fastest!
We're launching a $100,000 challenge to see who can make Muse Spark run fastest: openkernelchallenge.dev/
been thinking about why fast type checkers matter more for agents than for us
I run tsc a few times an hour. an agent runs it after basically every edit, reads the errors, tries again
so the checker's speed kinda is the agent's speed
say an agent checks 50 times in one task. that's ~12 mins sitting around on the old tsc vs ~20s on tsrs
and that's before linters, tests and builds, which agents hammer too