Shivam Kumar

If you doubt not. And fear not. Then you will know. That knowledge will be your salvation. L/0

[Artificial Analysis @ArtificialAnlys]

SpaceXAI just released Grok 4.5, and it ranks #4 on GDPval-AA v2 with an Elo of 1543 - behind only the latest Claude releases from Anthropic on real-world agentic knowledge work tasks

Grok 4.5 achieved this score at a cost of $0.49 per GDPval task to sit clearly on the Pareto frontier for performance versus cost. This cost is lower than GLM-5.2 and Kimi K2.6, and nearly 90% cheaper than the models ahead of it on our leaderboard.

We’re finalizing the remaining Artificial Analysis Intelligence Index evaluations and will share final results soon.

Thanks to @SpaceXAI and @elonmusk for their collaboration testing this model ahead of release, and congratulations on the launch!

[John Ennis @johnennis]

This attitude, if rampant at Google, helps to explain a lot

[Robbyant @robbyant_brain]

🌍 10 minutes was just the beginning. What if a world model could run forever — responding to your every action, generating new events, and never losing coherence?

Today, we open-source LingBot-World 2.0 (Infinity) 🔥

The first interactive world model with:
🔷 Hour-long generation with zero quality drift
🔷 Rich actions & events — attack, cast spells, shoot, summon storms
🔷 Agentic world — a Director Agent drives real-time world evolution

720p/60fps. Playable like a game. 🎮 Try it now on @reactorworld — or deploy it yourself with SGLang day-0 support. @lmsysorg

#WorldModel #EmbodiedAI #OpenSource #Robotics

[fal @fal]

First part of our series on how we built one of the fastest image generation systems, and we're kicking off with running LLMs (for prompt expansion) at >1000 tok/s by leveraging a DSpark adapter we trained from scratch!

https://t.co/qGFugOwCPo

[Marktechpost AI @Marktechpost]

NVIDIA Releases Audex (Nemotron-Labs-Audex-30B-A3B): A Unified Audio-Text LLM That Preserves the Text Intelligence of Its Backbone

Most unified audio models pay a text tax. Add audio output, and reasoning benchmarks drop — even when the only new output is speech. NVIDIA just released one that doesn't.

Audex (Nemotron-Labs-Audex-30B-A3B) is a 30B MoE with 3B active parameters, built on the text-only Nemotron-Cascade-2-30B-A3B backbone. Audio inputs are projected into the text embedding space. Text tokens and quantized audio tokens are then generated the same way, inside one MoE decoder — no thinker–talker split, no stacked cascade.

Here's what's actually interesting:

→ One model, audio in and out: understanding, ASR, translation, TTS, text-to-audio, speech-to-speech

→ Text holds vs its own backbone: IMO AnswerBench 81.1 vs 79.3, MMLU-Redux 86.4 vs 86.3

→ Beats text-only Qwen3.5-35B-A3B on several tasks: LiveCodeBench v6 85.3 vs 74.6, IFBench 77.8 vs 70.2

→ The usual tax, for contrast: Qwen3-Omni-30B-A3B-Thinking drops to 60.4 on HMMT vs 71.4 for its text backbone

→ 6.82 WER on OpenASR, ahead of Step-Audio-R1.1-33B (7.91) and Qwen3-Omni-Thinking (8.00)

→ Two codecs: X-Codec2 for speech (50 tok/s, FSQ, 65,536 codebook), X-Codec for general audio (200 tok/s, 4 flattened RVQ layers) — the only strong open model generating general audio beyond speech

Full analysis: https://t.co/UivFSXjsPU

Paper: https://t.co/AIZmA1RBcv

Model weights: https://t.co/Ve5h707tA1

@NVIDIAAI @nvidia @_weiping

[Vitto Rivabella @VittoStack]

🚨 Introducing: WallBreaker V1 🚨

An open-source AI red teaming CLI to help you research LLM jailbreaks and security.

- Probe LLMs guardrails
- Harmbench goals ready
- Find universal jailbreaks
- Fully autonomous or assisted campaigns
- Learns and improves after every successful run
- Computer use and MCP ready for live API testing

Set your attacking model, a target, select a goal, and you’re good to go.

WallBreaker will start probing different techniques and combinations based on its learnings and hundreds of data points until it succeeds.

This is the first open source tool coming out of the Jailbroken community.

⚠️ DISCLAIMER: For authorized use only. Point it only at systems you own or have explicit written permission to test. Unauthorized access can be a crime. Shipped as-is under AGPL-3.0: no warranty, no liability, zero endorsement of misuse.

Link in the comments 👇

[Gal Shir @galshirart]

It’s over. I’m quitting design.

A client of mine just created a logo with Fable 5, and the result left me speechless.

It understood the brand story, values, audience, strategy, and turned all of it into a smart, minimal symbol. A genuinely brilliant concept. The kind of idea that captures everything at once. Something I honestly don’t think I would have come up with myself.

And it didn’t just nail the idea. It executed the design pixel-perfectly.

So I raise the white flag.

My skepticism about AI’s ability to do great design is officially gone.

There, I said it:

AI beat me at design.

Now that AI finally took my job, I can peacefully quit and dedicate my life to studying the only thing it may never achieve: human consciousness and the pathways to God.

Good luck everyone.

[LangChain @LangChain]

We tuned the harness for @NVIDIAAI Nemotron 3 Ultra.

Benchmark-leading performance. 10x lower inference costs.

✅ An aggregate score of 0.86 at a cost of $4.48
✅ The closest-performing model: $43.48

https://t.co/xQmGdn8YPq

[Shivam Kumar @ShivamKumar212]

@AlfieJCarter AGENTS

[Siddhartha Saxena @siddsax]

Fable was so good, OpenAI had to assemble Luna, Terra, and Sol.

[Netflix Anime @NetflixAnime]

Fight to take back the future. Fight to end this fate.

Available August 5 on Netflix in select regions (🇩🇪🇦🇹🇨🇭🇱🇮🇱🇺):
★ Fate/Grand Order THE MOVIE Divine Realm of the Round Table: Camelot Wandering; Agateram
★ Fate/Grand Order THE MOVIE Divine Realm of the Round Table: Camelot Paladin; Agateram
★ Fate/Grand Order Absolute Demonic Front: Babylonia
★ Fate/Grand Order Final Singularity Grand Temple of Time: Solomon
★ Fate/stay night [Heaven's Feel] I. presage flower
★ Fate/stay night [Heaven's Feel] II. lost butterfly
★ Fate/stay night [Heaven's Feel] III. spring song

[Elon Musk @elonmusk]

Our internal assessment is that Grok 4.5 is roughly comparable to Opus 4.7, but much faster. The combination of capability, faster speed and lower cost is what makes it competitive.

We are closing the loop on real-world usefulness, not benchmarks. Hardcore engineers at Tesla & SpaceX find Grok 4.5 genuinely useful, which is what actually matters.

[Cursor @cursor_ai]

We've partnered with SpaceXAI to train Grok 4.5.

It’s our most powerful model yet and the first we've built for more than software engineering.

[Robbyant @robbyant_brain]

Today we open-source LingBot-Video — the first MoE-based video foundation model built for embodied intelligence.
🔹30B params, only 3B active at inference.
🔹Augmented with 70K hours of embodied data on top of large-scale internet video pretraining.
🔹Already outperforming Wan2.6, Seedance 1.5 Pro, and Cosmos3 Super on RBench.
🧵👇

[BridgeMind @bridgemindai]

GROK 4.5 JUST WENT LIVE IN CURSOR.

256k context. Built for long running agentic coding.

Elon's claim: Opus 4.7 level, but faster and cheaper.

Testing that claim right now on real production code.

BridgeBench run starts today.

If this actually holds Opus level quality at Grok speed and Grok pricing, the economics of AI coding change this week.

[Cognition @cognition]

Introducing SWE-1.7, the most capable model we’ve trained yet.

It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s.

RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

[BridgeMind @bridgemindai]

SWE 1.7 is a scam.

I prompted it ONCE. One single prompt.

It dumped 3,477 lines across 12 files, exhausted my entire daily quota, and every line of it was slop.

It is genuinely the fastest model I have ever seen. At generating garbage.

The Slopinator.

Speed is not the metric.

Slop per second is not intelligence.

Burning my whole quota in one prompt is not a feature.

[BridgeMind @bridgemindai]

New CursorBench results just dropped and Grok 4.5 is the story.

#3 overall at 66.7%. Right behind Fable 5 Max at 70.5%.

Now look at the cost column.

Fable 5 Max: $17.32 per task
Grok 4.5 High: $1.51 per task

That is Fable level performance at roughly 1/10th the cost.

And it beats Fable 5 High and Opus 4.8 Max outright.

The intelligence war just became a price war.

[OpenAI @OpenAI]

Introducing GPT-Live, a new generation of voice models for natural human-AI interaction.

Rolling out in ChatGPT starting today.

You’ll want to turn the sound on for this one.

[🍓🍓🍓 @iruletheworldmo]

i’ve been using sol for two months so i’ll write a longer post later but it’s basically a hair below fable for intelligence, has slightly worse design taste.

but you’ll be able to use it a hell of a lot more and it’s a lot better for longer horizon agentic work. far more steerable.

i have some words to say however about how i feel about getting such an improved model two months before others. not entirely positive words.