Codacus

[Michael Helcig @mhelcig]

Our largest model yet.

763B parameters, compressed with GSQ-RCO and co-optimized with Strata (by @coldniko) for local inference.

On Hugging Face this Friday.

Much more is coming. Follow us for future releases.

[Codacus @thecodacus]

@mehulmpt "dont blink" - Dr. Who

[Codacus @thecodacus]

@coldniko @murasametech love it

[Codacus @thecodacus]

when are you going to support deepseek 🙂

[Niko Veit @coldniko]

Beta testers so far loving it - just annoyed with few bugs 👀

[Codacus @thecodacus]

@2ndGenAmerica depends on what that is.. if you want to simulate things in virtual environments, like physics or lights simulations, in science then you will need coding to implement it or to verify results from papers.

but other than that I haven't used them for science or math yet.

[Codacus @thecodacus]

Wow the other side is now, entering the arena,
77 on DeepSWE and Peaked in Automations!!
waiting for others to show their cards

[DeepSeek @deepseek_ai]

🚀 DeepSeek-V4.1-Flash-1017 is here!

Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.

🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.

1/4

[Codacus @thecodacus]

Hot Take: JEV is just an LLM with one single pass parallel call for all the field..

[Diogo Almeida @CompleteSkeptic]

Jev was multimodal all along

Yep, the Jev you know and love was an image model this whole time. We just chose to disable it because models tend to be less intelligent on other modalities.

Header: TypeSafe-Danger-Zone-Images: true

[Codacus @thecodacus]

haha.. looks like we have a naruto fan here 😉

[Diogo Almeida @CompleteSkeptic]

don't watch how the fight ends tho

[Codacus @thecodacus]

LOL

[Mehul Mohan @mehulmpt]

"We used our models to understand the device and where it had come from"

[Codacus @thecodacus]

@levelsio this is really awesome

[Codacus @thecodacus]

This is exactly what I was thinking of building for a long time.. will this work with any engine?

Thanks for doing it 😁

[石ちゃん🎤LLMでアプリ開発 @Isichan_Hitori]

【 Angio 開発画面 】
～ Strataのボトルネック可視化ツール～

Strata GLM 調整の時、遅い個所の把握が大変でした💦
なので血管造影（Angio）みたいに、流れ道の「つまり」を見るアプリ作ってます‼

・ボトルネックの場所が分かる
・直す価値があるかどうかがわかる

近日中にリリース予定です🌟

[Codacus @thecodacus]

So finally we have a sub 40b moe model other than the qwen

[OpenBMB @OpenBMB]

🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series!
Advancing the Performance–Efficiency Pareto Frontier.
It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency.
Two new models:
🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to local devices with newly introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI.
🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy.

⚡ What's new?
SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension.
Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯
Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B.
Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance.
Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS!

Download, deploy, and build something we haven't imagined yet. 👀👇
🔗 GitHub: https://t.co/msjm2QlrvX
🤗 Hugging Face: https://t.co/W95QKGhymw

[Codacus @thecodacus]

@mehulmpt btw what was the prompt?

[Codacus @thecodacus]

@bijanbowen Lol, no Sero has been harmed

[Codacus @thecodacus]

Here is another one with
Qwen3.8 Flash Next : IQ3_S

[Codacus @thecodacus]

Here is one from me.. since this is a trend now
qwen 3.8 flash q3 on 12GB gpu 64GB ram

[Codacus @thecodacus]

but for that to happen, the people who do know how to run it has to make it seemless

[Matthew Berman @MatthewBerman]

Local AI will work when people don't know it's running locally. It has to be that seamless.

[Codacus @thecodacus]

Since everyone is running glm now.. here is mine
GLM-5.3-Flash, a 321B model, running on two RTX 3060s (24 GB VRAM total).

13 tokens/s. 325 t/s prompt reading.

Engine: Project Maya. Quant: Maya S24

[Yann LeCun @ylecun]

@BrianRoemmele @elonmusk The Medal of Science is for scientists.
Scientists publish their works in peer-reviewed venues so they can be scrutinized, verified, and reproduced.

[Codacus @thecodacus]

@simonw run qwen3.8 flash next 125B, at Q3 at 40t/s and 1000+t/s prefill, with 60GB ram, with RTX 3060 12GB vram.. and a fast ssd, with strata, its a no brainer at this point