███

Ragebaiting account; tweets will be used for SFT.

[███ @benchmaxxed]

Here’s the truth: There is no “new tokenizer.”

Anthropic needs a way to secretly raise prices. What’s the best way to do that? Simple.

Introduce a “new tokenizer” that eats tokens like never before. Free money for 0 extra compute.

Anthropic: the most evil company in the world.

[wh @nrehiew_]

The 4.7 tokenizer treats whitespace as separate tokens?

A string consisting of 50 one-token words separated by Whitespace tokenizes to ~50 more tokens than with the 4.6 tokenizer.

If so, the 1.35x more token estimate seems way too low.

[███ @benchmaxxed]

Remember when some retard @yegor256 suggested replacing all API endpoints with a LLM?

I wonder how many other LLM illiterate retards with AI psychosis are out there, ordering their engineers to implement similar features.

[impulsive @weezerOSINT]

meta gave their AI support agent the ability to modify your instagram account. no identity verification. people figured this out and accounts are being taken over right now

[impulsive @weezerOSINT]

meta gave their AI support agent the ability to modify your instagram account. no identity verification. people figured this out and accounts are being taken over right now

[Elon Musk @elonmusk]

@eliebakouch We will open source the 0.5T model towards the end of this year. It should still be quite useful.

[DeepSeek @deepseek_ai]

We are making our discount permanent! 🎉

Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀

[███ @benchmaxxed]

@antirez 2.5 and 2.7 are just 230B in size; you can only store a limited amount of precise coding knowledge with such few parameters.

The models score high because MiniMax is known to benchmax. Except for cheaply running Hermes agents, MiniMax models are practically useless.

[███ @benchmaxxed]

@aravindr93 Performance compared to Qwen 3.5, which I assume is your base? How much value does your fine tune add?

[███ @benchmaxxed]

@morqon Source?

“I asked ChatGPT.
No I did not do any calculations.
No I did not ask for calculations.
No I did not ask for assumptions.
No I don’t understand what was said.
But I’m gonna make a bold claim on Twitter anyways.”

The q in morqon is silent isn’t it?

[███ @benchmaxxed]

@ar0cket1 Prices? Margins?

You realize that DeepSeek V4 Pro is served at only 3.5/1.75 with 4x the parameters of your Gemini 3.5 Flash estimation right?

I agree that the benchmark is unreliable, but there is almost certainly no way Flash is not at least a 1T model.

[███ @benchmaxxed]

@ar0cket1 I strongly doubt it’s a 400B model or smaller based on the amount of world knowledge it has. But who knows, maybe Google found a hidden data compression technique. (Judging by Gemma 4’s world knowledge, likely not)

See:
https://t.co/ifbBUY2JOZ

[███ @benchmaxxed]

@astrogu_ How is this a surprising result? You obviously lose performance if important information is not inside the context. That’s the point of an AGENTS.md file.

[███ @benchmaxxed]

Benchmaxxed.

[Mercor @mercor_ai]

Gemini 3.5 Flash tops the APEX-Agents AA leaderboard at 47.1%, benchmarked by @ArtificialAnlys.

That's nearly 10 points ahead of GPT-5.5 (37.7%) and 19 points above Gemini 3 Flash (27.7%).

Full APEX-Agents run coming soon 👀

Check out the APEX-Agents-AA leaderboard: https://t.co/uzIS2NsXBA

[███ @benchmaxxed]

*Kimi K2.6

[Michael Truell @mntruell]

Composer 2.5 is a significant step up from Composer 2.

This is the very start of our work with SpaceXAI. Hope to have more improvements out soon.

[Everlier @Everlier]

@ThePrimeagen It's called "zero" because LLMs have zero latent knowledge of this language, everything will have to be spoon fed from our beloved markdown files

[███ @benchmaxxed]

@Muennighoff *Kimi K2.6

[███ @benchmaxxed]

@JackKuhr “Third most ARR”

Great. Now let’s see the operating costs.

[BuccoCapital Bloke @buccocapital]

There’s nothing to reinvent. You pick up a book. You read the book. What are we doing here? Stop it

[███ @benchmaxxed]

Your daily reminder that C suite MBAs are all retards who don’t know shit about tech

[Yegor Bugayenko @yegor256]

RESTful APIs may be dead soon. Instead, web services may expose a single POST entry point for a prompt. Internally, an AI agent may decide how to interpret it and what to do with the data and the database.

[███ @benchmaxxed]

Translation: please ban open models and let us continue charging $5/25 for our shitty models

[Anthropic @AnthropicAI]

We've published a paper that explains our views on AI competition between the US and China.

The US and democratic allies hold the lead in frontier AI today. Read more on what it’ll take to keep that lead: https://t.co/TgJBeodWYK

[███ @benchmaxxed]

Only 35k but it’s understandable because it’s Midstral. Who the fuck even wants data from a company who’s 2 years behind in the AI race?

[Dark Web Informer @DarkWebInformer]

‼️🇫🇷 Mistral AI allegedly breached: ~5GB of internal source code and ~450 private repositories exposed from the French AI company by TeamPCP

A threat group is selling approximately 5GB of internal repositories and source code allegedly belonging to Mistral AI and Mistral Solutions, covering training, fine-tuning, benchmarking, dashboard/platform, model delivery and inference, experiments, and future projects.

The actor is demanding a $25,000 BIN, stating they will shred the data permanently and sell to one buyer only, and threatening to leak all ~450 repositories for free to the forums within a week if no buyer is found.

▸ Actor: TeamPCP
▸ Sector: Artificial Intelligence / Source Code
▸ Type: Data Sale (with leak threat)
▸ Records: ~450 internal repositories, ~5GB total
▸ Country: France
▸ Date: 11/05/2026

Compromised data:

▪ mistral-inference-internal.tar.gz
▪ mistral-inference-private.tar.gz
▪ mistral-lawyer-internal.tar.gz
▪ mistral_finance_agent.tar.gz
▪ mistral-compute-poc.tar.gz
▪ mistral-fabric.tar.gz
▪ finetuning-feedback.tar.gz
▪ mistral-finetune-internal.tar.gz
▪ cma-customer-care-internal.tar.gz
▪ mistral-common-internal.tar.gz
▪ chatbot-security-evaluation.tar.gz
▪ kyc-doc-agent.tar.gz
▪ dashboard.tar.gz
▪ devstral-cloud.tar.gz
▪ finance.tar.gz
▪ typhoon.tar.gz
▪ turbine.tar.gz
▪ mistral-surge.tar.gz
▪ mistral-solutions.tar.gz
▪ surge-validators.tar.gz
▪ website-v3.tar.gz
▪ xformers.tar.gz
▪ piper-segmentation.tar.gz
▪ pfizer-rfp-2025.tar.gz
▪ Internal repositories tied to model training, fine-tuning, benchmarking, dashboard and platform code, model delivery and inference systems, experiments, and future project work

Stop guessing what's redacted. Subscribers see everything → https://t.co/281Qjc6p2J