RELOX AI

I keep an eye on AI so you don’t have to. Tracking the latest AI news, LLM updates, and practical tools daily.

[RELOX AI @Relox_AI]

AI safety isn't a wall around the knowledge.

It's a habit painted on top.

Here's how jailbreaks really work, explained in 2.5 minutes.

[RELOX AI @Relox_AI]

Google DeepMind's models are helping make disease more detectable, treatable, and preventable. From pinpointing rare genetic variants to speeding up drug discovery. Lab breakthroughs turning into real-world care. #Google #AI

[Google DeepMind @GoogleDeepMind]

Together with scientists around the world, our models and tools are helping make disease more detectable, treatable, and preventable. 🧬

From pinpointing rare genetic variants to speeding up drug discovery, here’s how we’re helping turn lab breakthroughs into real-world care → https://t.co/sM5G0CLcKJ

[RELOX AI @Relox_AI]

Tesla's Roadster just did a kickflip. An actual aerial flip, mid-air. The flying was already insane, now it's doing stunts. #Tesla #AI

[🚨 AI News | TestingCatalog @testingcatalog]

Imagine testing this for the 1st time 🥵

[RELOX AI @Relox_AI]

Wow this is insane. 🔥

[Joss Belmont @belmoncode]

Jev @typesafeai can literaly save lives!! made it detect drowning from video footage. It watches over your kids at the pool for less than $2.5/hr

[RELOX AI @Relox_AI]

Tesla's Roadster flies. Not a render, not CGI, the actual car hovering in the air. 0-60 mph in 0.81 seconds. We really live in the future now. #Tesla #AI

[Tesla @Tesla]

Real life, baby

[RELOX AI @Relox_AI]

StepFun just dropped Step 5 Preview as open weights. 600B total / 27B active sparse MoE, 1M context, native text and vision input. BF16 and FP8 live on HuggingFace and ModelScope. Frontier intelligence, open weights. #StepFun #AI

[StepFun @StepFun_ai]

Step 5 Preview is now open weights.

BF16 and FP8 weights are live on Hugging Face and ModelScope, together with our technical blog.

• Sparse MoE: 600B total, 27B active per token
• 1M-token context window
• Native text and vision input
• Built for long-horizon agentic work: coding, software engineering, professional knowledge work and finance

Frontier intelligence, open weights. Build what's next.

Hugging Face: https://t.co/nMfayf37ja
ModelScope: https://t.co/mkB3qIoMQA
GitHub: https://t.co/siRkQqaruX

[Elon Musk @elonmusk]

Turns out the future does have flying cars

[RELOX AI @Relox_AI]

@Y_Jehoash wow this is crazy 😮

[RELOX AI @Relox_AI]

Anthropic just gave @METR_Evals employee-like access to assess its safety practices. They're in talks with more orgs too. Transparency on evals is the right move. #Anthropic #AI

[Anthropic @AnthropicAI]

We’ve committed to having evaluators assess our safety practices. We're in discussions with multiple organizations, and have now given employee-like access to a team from @METR_Evals.

In this post, we describe how we work with third-party evaluators: https://t.co/8CLQnMVg4X

[RELOX AI @Relox_AI]

Microsoft's MAI-Base-2 pre-training is done, and MAI-Base-3 is already in prep. 3T-parameter class, built entirely in-house. 4-12x more FLOPs-efficient than the strongest open base model. Post-training results coming soon. #Microsoft #AI

[🚨 AI News | TestingCatalog @testingcatalog]

MICROSOFT 🔥: MAI-Base-2 pre-training has been completed; MAI-Base-3 training cycle is already being prepared as well.

> "MAI-Base-2 outperforms every open-weight model we tested on our held-out evaluations. It is 4-12x more pre-training FLOPs-efficient than the strongest open base model, DeepSeek-V4-Pro-Base (see Figure 3 below), and we have put that saving into training a larger model."

* Figure 1 shows the performance of MAI-Base-2 and other models with at least one trillion total parameters across a range of pre-training evaluations (lower is better).

[RELOX AI @Relox_AI]

Black Forest Labs just launched FLUX 3 Image Klein, a new open-weight image model. 5.6B params, runs on a single consumer GPU. Native 4K, bounding boxes, pixel-perfect editing. Customizable and fast. #FLUX #AI

[🚨 AI News | TestingCatalog @testingcatalog]

Black Forest Labs launched FLUX 3 Image Klein, its new open-weight image model!

> "Customizable and fast. 5.6B parameters and runs on a single consumer GPU. Native 4K, bounding boxes, and pixel-perfect editing."

> Developers can post-train and fine-tune FLUX 3 Image Klein.

> It comes with a low-latency version for faster generation and self-hosting, as well as quantized versions - FP8, NVFP4, and INT8.

> The distilled model with native 4K will also be available on BFL APIs.

[RELOX AI @Relox_AI]

Gemini 4 Argon is rolling out to Google AI Pro subscribers on Antigravity, with a dedicated weekly quota just for Argon. That was fast. #Google #AI

[🚨 AI News | TestingCatalog @testingcatalog]

GOOGLE 🔥: Gemini 4 Argon is rolling out to Google AI Pro subscribers on Antigravity!

It comes with a new dedicated weekly quota, specifically for Argon model.

That was fast! 👀

[RELOX AI @Relox_AI]

Meta's Omnilingual ASR just made TIME's Best Inventions of 2026. Speech recognition for over 1,600 languages, including 500 low-coverage languages never served by any ASR system before. A real step toward universal transcription. #Meta #AI

[AI at Meta @AIatMeta]

We’re proud to share that Meta Omnilingual Automatic Speech Recognition (ASR) has been named one of @TIME’s Best Inventions of 2026!

Meta Omnilingual ASR is a suite of models providing ASR capabilities for over 1,600 languages, including 500 low-coverage languages never before served by any ASR system.

While most ASR systems focus on a limited set of languages that are well-represented on the internet, this release marked a major step toward building a truly universal transcription system.

Learn more about it here: https://t.co/PPdTp8RIKt

[RELOX AI @Relox_AI]

OpenAI is expanding GPT-Rosalind access for academic research with a new Rosalind Academic Plan. Eligible US university labs get up to 12 months of sponsored access. Research data won't train their models by default. #OpenAI #AI

[OpenAI @OpenAI]

We're expanding access to GPT-Rosalind for academic research.

The new Rosalind Academic Plan helps scientific labs bring AI into the work of understanding biology, exploring ideas, and deciding what to test next—with sponsored access for eligible labs at U.S. universities.

Academic research starts with questions whose impact may take years to impact human health.

[RELOX AI @Relox_AI]

@chetaslua Damn bro, this is looking insane. Oh my god, Anthropic cracked something internally fr I just love Anthropic. 😍

[RELOX AI @Relox_AI]

OpenAI just shipped GPT-Live-1.1 for Voice Mode in ChatGPT, Codex, and dots. New voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese. Better accents, less 'checking...' Rolling out to everyone today. #OpenAI #AI

[OpenAI @OpenAI]

New voices. Better accents. And less "checking..."

GPT-Live-1.1 brings these updates to Voice in ChatGPT, Codex, and dots, with new voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese.

Rolling out today to all users. Coming soon to the API.

[RELOX AI @Relox_AI]

Meta just open-sourced Muse Spark: 816B total, 42B active, 1M context, natively multimodal. They're calling it the most capable US open-weight model to date. Runs on two Apple M5 Ultra Mac Studios. #Meta #AI

[AI at Meta @AIatMeta]

Today we’re releasing Muse Spark Open, the most capable US open-weight model to date.

Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning.

Key Model Specs:
→ 816B total / 42B active parameters
→ 1M token context window
→ Natively multimodal across text, image, video, audio

[RELOX AI @Relox_AI]

Krea AI just launched Krea 3, a new image model built for fast, precise editing. Krea 3 Ultra scored 4.81/5 on CPI-Bench, second only to GPT Image 2.5. 10x faster than GPT Image 2, from $0.03 per image. #Krea #AI

[🚨 AI News | TestingCatalog @testingcatalog]

Krea AI introduced Krea 3, a new image model designed for fast, precise image editing.

> Krea 3 combines separate references for characters, their outfits, and the location into a single scene while keeping faces and color grading consistent.

> On CPI-Bench, a benchmark for real-world image editing, Krea 3 Ultra scored 4.81 out of 5, second only to GPT Image 2.5 and ahead of every other image model.

Testing time! 🔥

[RELOX AI @Relox_AI]

It's official: NVIDIA just announced Nemotron 3.5 Super VL. Open model, 120B total with 12B active, built for reasoning and long-running agents with text, image, document, and video understanding. #NVIDIA #AI

[🚨 AI News | TestingCatalog @testingcatalog]

NVIDIA 🔥: A new Nemotron 3.5 Super VL model has been announced!

> NVIDIA Nemotron 3.5 Super VL is an open model with 120B total parameters and 12B active.

> The model supports reasoning, tool use with text, image, document, and video understanding for long-running AI agents.

[RELOX AI @Relox_AI]

OpenAI is putting $5 million into a new Simons Foundation math initiative. The goal: help mathematicians examine, understand, and build on AI-produced results. Simons is committing $15M total across OpenAI and Anthropic partnerships. #OpenAI #AI

[OpenAI @OpenAI]

We want to support the mathematics community in turning progress in AI into deeper human understanding and new discoveries. To support that goal, we are contributing $5 million to a new Simons Foundation initiative to help mathematicians examine, understand, and build on results produced by AI, including the results we recently published.

The funding, managed and awarded by the @SimonsFdn will support workshops, conferences, and special programs. We will also provide access to our models and computing resources to Simons Foundation-supported researchers.

[RELOX AI @Relox_AI]

Claude-Next is back, stealth-routed through Opus in Claude CLI and web. Could be Fable 5.5, Opus 5.6, or something else entirely. The unreleased model watch continues. #Anthropic #AI

[Kol Tregaskes @koltregaskes]

"Claude-Next" has returned to Claude CLI and Claude web via stealth routing through Opus. This could be Fable 5.5 or Opus 5.6 or something else.