RazerAI

Razer is leading a new era in AI-powered gaming—redefining how games are developed and experienced. We’re hiring:

[RazerAI @RazerAI]

Razer AIKit v0.8.0 is here. ⚡

This release brings System One decision models to AIKit, with two open-source model families ready to run locally:
→ Mapika Decider: 0.8B / 2B / 9B
→ AutoTrust Jev: 9B / 27B / 27B-VL

But instead of only benchmarking tokens/sec... We let
it play DOOM. 🔫

More results soon.

[RazerAI @RazerAI]

No cloud. No API keys. No meter running.

Razer AIKit is our open-source toolkit for running LLMs and decision models on your own single or cluster of GPUs, served by vLLM through an OpenAI-compatible API: https://t.co/NSCgUhrrOX

#RazerAIKit #LocalAI #AIDevelopment

[Min-Liang Tan @minliangtan]

Local models are where this gets fun. The cloud will keep training the big brains, but the AI you use every day should run on the device in front of you. Private, fast, no meter running. https://t.co/DulBxCy88o

[Min-Liang Tan @minliangtan]

Been using Grok Bot and it's pretty amazing.

Got my team to hack together a GrokPuck (Grok Bot in your pocket). Swipe the screen, the bot changes.

What do you guys think?

[RazerAI @RazerAI]

We put three AI models behind the controls of DOOM and compared what happened in real time.

At the same model size as Qwen, Mapika delivered 4x the kills.

→ Mapika Decider 0.8B, local: 40 ms per decision, 12 kills
→ Qwen 3.5 0.8B, local: 90 ms, 3 kills
→ TypeSafeAI/Jev via cloud API: 270 ms, 5 kills

But the interesting part isn't just the score. It's the timing.

The cloud model made strong decisions, but at 270 ms, they often arrived too late. In a real-time environment, even a good decision loses value if the moment to act has already passed.

Running locally also removes per-call API costs, rate limits, and connectivity dependencies.

See the full results and what we learned with Razer AIKit v0.8.0: https://t.co/eJpGYnAZ2M

#RazerAIKit #LocalAI #AIDevelopment

[RazerAI @RazerAI]

What does it take to run decision models locally?

Explore the DOOM test and what’s new in Razer AIKit v0.8.0, including support for JEV-style decision models.

Read the article: https://t.co/KZoHnmZAa3

#RazerAIKit #LocalAI

[RazerAI @RazerAI]

Attack or retreat. Route A or B. Escalate or not.

For many AI agents, the job is to choose. So why hand those decisions to models built to talk?

We tested a local decision model in DOOM. See how it played and what we learned. Article link in the replies. 👇

[RazerAI @RazerAI]

Every part of Barracuda AI was designed with that in mind.

Stay tuned to find out: https://t.co/xmhbzjl75i

#BarracudaAI #AIWearables #AmbientAI

[RazerAI @RazerAI]

The loudest question in AI wearables right now: should a device record your life before you’ve asked it anything?

Our view: ambient shouldn’t mean always on the record.

Privacy isn’t a feature. It’s a design principle.

[RazerAI @RazerAI]

We said more results soon. Here they are. 🔫

One DOOM run, three models.

A local decision model, a local LLM and a cloud API. Only one made its calls in time.

See who won and by how much: https://t.co/KZoHnn07ZB

#RazerAIKit #LocalAI

[RazerAI @RazerAI]

Razer AIKit v0.8.0 is here. ⚡

This release brings System One decision models to AIKit, with two open-source model families ready to run locally:
→ Mapika Decider: 0.8B / 2B / 9B
→ AutoTrust Jev: 9B / 27B / 27B-VL

But instead of only benchmarking tokens/sec... We let
it play DOOM. 🔫

More results soon.

[RazerAI @RazerAI]

Ever tried asking for help while standing on a train platform?

By the time you've explained the situation and what you need, your train could already be gone.

That's the gap perceptual intelligence closes: AI that draws on what you see and hear, so you don't have to explain every detail from scratch.

“Is this the right train?” should be all you need to ask.

#RazerBarracudaAI

[Hien | Commissions OPEN | @HienKong]

@Razer just revealed their latest companion Sumi! Team Razer reached out for an awesome collab idea they had in mind for me. So here is Sumi facing off against Ryu using their Razer Kitsunes!

#PixelArt #ドット絵

[R Λ Z Ξ R @Razer]

Every detail tells a part of Sumi's story: https://t.co/6R28ZLUt37

Go behind the scenes with the team that brought Sumi to life, from her earliest character concepts and digital designs to 3D modeling, motion capture, and the creative touches that make her who she is.

[R Λ Z Ξ R @Razer]

Meet Razer AVA, with Sumi at the heart of the experience: https://t.co/6R28ZLTVdz

Your next chapter in AI companionship starts here.

[RazerAI @RazerAI]

For 20 years, we’ve believed everyone will be a gamer. AI could bring that future closer, making gaming more personal, creation more accessible, and development more efficient.

In a recent interview with @FortuneMagazine, our CEO Min-Liang Tan shares how AI could help gaming become a trillion-dollar industry, with innovations across our AI pipeline, from Razer Barracuda AI and Razer AVA to QA Companion-AI, reshaping how we experience, create and develop games.

Read more: https://t.co/XrHSWHRyiC

#Razer #AI #Gaming #ForGamersByGamers #Innovation

[RazerAI @RazerAI]

View the latest release on GitHub: https://t.co/Y3ncojswsD

Highlights:
1️⃣ System One Decision Models 🧠
2️⃣ vLLM v0.30.0 Engine Upgrade ⚙️

[RazerAI @RazerAI]

Razer AIKit v0.8.0 is here. ⚡

This release brings System One decision models to AIKit, with two open-source model families ready to run locally:
→ Mapika Decider: 0.8B / 2B / 9B
→ AutoTrust Jev: 9B / 27B / 27B-VL

But instead of only benchmarking tokens/sec... We let
it play DOOM. 🔫

More results soon.

[RazerAI @RazerAI]

It also has to earn its place on your body: comfortable for hours, obvious when it's on, easy to switch off.

Form factor decides what the AI can perceive, and whether you keep wearing it.

What would it take for you to wear AI all day? 🤔

#RazerBarracudaAI

[RazerAI @RazerAI]

Ambient AI isn't AI that's always watching. It's perceptual intelligence that's ready when you are, with the context already in hand.

Most wearables pick one sense: glasses see, earbuds listen. Context needs all of it: what you see, what you hear, and what you've already said.

[R Λ Z Ξ R @Razer]

“Somewhere in this world, there's a place for me... isn't there?” – Sumi

Tune in for the full reveal on Oct 20.

[RazerAI @RazerAI]

Search by context. Keep your data local. 💻

Razer AIKit brings local semantic search + RAG to your machine, grounding AI answers in your documents with complete privacy.

Try the notebooks at https://t.co/Y3ncojswsD 📒

#LocalAI #RAG

[RazerAI @RazerAI]

AI has evolved. The way we access it hasn’t.

Models can now reason across text, images, audio, and code. But using them is still largely manual:
1⃣ Unlock phone
2⃣ Open an app
3⃣ Capture or type the context
4⃣ Send it and wait for response

The problem is context. AI still depends on people to translate the world around them into prompts.

The next step is AI that can understand what is in view and respond in context, with fewer steps in between. Less time explaining, more direct interaction with the world around us.

Be the first to experience AI beyond the screen: https://t.co/Pd7TCmpgIg