Digi Decode

Follow for AI & Tech related updates, developing

[Digi Decode @digi_decode]

@shuai_bai_ Qwen's vision is always remarkable, version 2 used to use a ton of memory but not anymore

[Digi Decode @digi_decode]

@teortaxesTex I don't know but it seems to be Alibaba Cloud stress testing their newly delivered Atlas 950 SuperPoDs, the 100t is likely calculated on compute bases and not memory
The model is likely 80b8a or 122b10a or something similar

[Digi Decode @digi_decode]

Ox Alpha is likely a Qwen MoE model and will debut next week, 100T daily token capacity goes with the model being MoE and backed by large infra (Alibaba Cloud)

[OpenCode @opencode]

Ox Alpha (stealth model) is free for the next week

- 1M Context
- Multi-modal
- Zero Data Retention

Generous rate limits, near unlimited usage

We have capacity for 100T tokens per day, lets see what you can do

[Digi Decode @digi_decode]

@bnjmn_marie I would say purpose of benchmark is to also measure adaptability, so 76.2% would be actual score, but in a real-world task the user would likely install Numba to help it more forward like you did.

[Digi Decode @digi_decode]

@IntCyberDigest Could have just unplugged it 😅

[Digi Decode @digi_decode]

DeepSeek v4 Flash 0731
- Scores 52 in AA
- 13b active params

Qwen 3.8 27b
- Scores 52 in AA
- 27b active params

Does that mean future Qwen 27b still has more to give probably even Fable 5 level score ...

[Digi Decode @digi_decode]

Qwen 3.8 27b almost as good as Qwen3.8 Max🤯🤯on Vals Index v2

[Vals AI @ValsAI]

Qwen 3.8 27B is #6 among open-weight models on the Vals Index v2 and is #1 on Harvey's Legal Agent Benchmark among open weights.

[Digi Decode @digi_decode]

@NielsRogge Give SlopOn a try
https://t.co/qjl8GSbVBX

[Digi Decode @digi_decode]

@scaling01 here Qwen 3.8 27b zero shot the same prompt, and that with UD-Q4_K_XL quant not even full precision, it didn't work for you maybe because of the harness you used

[Lisan al Gaib @scaling01]

literally just tried to let qwen3.8 27b one shot a fluid simulation webapp

it thought for 40k tokens, took around 1 hour to generate and it's just black and doesn't work

meanwhile Opus 4.5 just oneshots it in a minute and works wonderfully

exact same prompt

[Digi Decode @digi_decode]

GLM 5.3 on par with Kimi K3
K3 has about 3.7x more total parameters
K3 has around 2.6x more active parameters

[Lou @louszbd]

A month ago, we asked for prompts that models still struggled with. We received many thoughtful examples.
Today, GLM-5.3 reaches 60 on the Artificial Analysis Intelligence Index, with 743B base.
Thank you to everyone who contributed.

[Digi Decode @digi_decode]

@saranormous My harness sits at 2 stars, 9998 more stars to go
https://t.co/qjl8GSctrv

[Digi Decode @digi_decode]

That reads like a threat to all the corporations with internet presence 😬
Pay now or face consequences 😅

[Greg Brockman @gdb]

defenders can see the future, and have a narrow window to uplevel their cybersecurity practices now.

key is to uplevel fundamentals and apply the best AI tools.

what we’re doing at OpenAI, and where other organizations can start: https://t.co/P3IMZkV234

[Digi Decode @digi_decode]

@sarahookr Given the amount of money thrown around, I doubt any breakthrough would get published publicly, sadly.

[Digi Decode @digi_decode]

Much better outcome for OpenRouter customers than being acquired by OpenAI which was rumoured.

[Andrew Curran @AndrewCurran_]

Stripe has finalized an agreement to acquire OpenRouter for more than $7 billion, according to a report from Bloomberg. This is more than five times the $1.3 billion valuation from OpenRouter's funding round just 82 days ago.

[Digi Decode @digi_decode]

@addyosmani Like Patrick, I too don't get the terminal-based coding harnesses, I quite liked Roo code but that is only single session, so I built https://t.co/fcNfCFwxOn , it is multi session, multi project, multi repo, I find it help with focus on task instead of waiting for LLM response.

[Digi Decode @digi_decode]

Suno level music generation on your local machine, no subscription needed, unlimited generations

[MiniMax (official) @MiniMax_AI]

🎵MiniMax-Music3
Next-Generation Open-Weights Production-Ready & Versatile Music Model

https://t.co/V2rxZk4xvh

[Digi Decode @digi_decode]

@ThePrimeagen since you would be generating more slop why not try https://t.co/47VAoD86jm

[Digi Decode @digi_decode]

Great model another win local AI but Qwen 3.8 27b will be out in few days

[AI at Meta @AIatMeta]

Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows.

Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on consumer hardware like a Mac or PCs with performant GPUs.

In keeping with our long tradition of sharing fundamental AI research, we’re releasing model weights under a permissive Apache 2.0 license.

🧵👇

[Digi Decode @digi_decode]

@SullyOmarr it's an infinite slop generator, give it a spin