- Horizon AI
- Posts
- xAI launches Grok 4 Fast ⚡
In partnership with
Welcome to another edition of Horizon AI,
Elon Musk’s xAI has introduced a new version of Grok that not only significantly reduces costs but also maintains the high performance of its previous model.
Let’s take a look!
Read Time: 4.5 min
Here's what's new today in the Horizon AI
xAI Launches a Faster and More Cost-Effective Version of Grok 4
Luma Debuts New Video Generation Model With "Reasoning" Capability
AI Tutorial: How to turn sketches into images with Grok
AI Tools to check out
AI Findings/Resources
The Latest in AI and Tech 💡
AI News
XAI
xAI Launches a Faster and More Cost-Effective Version of Grok 4
xAI launched Grok 4 Fast, a lighter model that delivers performance close to Grok 4 while being significantly more efficient, with some tasks costing up to 98% less according to xAI.
Details:
The new model scores 85.7% on GPQA Diamond and 92.0% on AIME 2025, achieving performance close to Grok 4 and GPT-5.
Grok 4 Fast achieves this while using 40 percent fewer thinking tokens on average than Grok 4. The gap becomes more noticeable on complex problems, where other models take more intermediate steps and require more computation.
It also outperforms Grok 4 on BrowseComp (44.9%) and X-Bench DeepSearch (74%), and tops OpenAI’s o3-websearch on LMArena-Search.
The model comes in two versions, one optimized for reasoning-heavy work and the other for quick answers, both supporting a 2-million-token context window.
Grok 4 Fast is available via grok.com, mobile apps, and the xAI API, with pricing from $0.05 to $1.00 per million tokens. For now, it is also free to use through OpenRouter and Vercel.
TOGETHER WITH GUIDDE
Create How-to Videos in Seconds with AI
Stop wasting time on repetitive explanations. Guidde’s AI creates stunning video guides in seconds—11x faster.
Turn boring docs into visual masterpieces
Save hours with AI-powered automation
Share or embed your guide anywhere
How it works: Click capture on the browser extension, and Guidde auto-generates step-by-step video guides with visuals, voiceover, and a call to action.
LUMA AI
Luma Debuts New Video Generation Model With "Reasoning" Capability
Luma AI launched Ray3, a new video generation model that can produce studio-quality HDR videos and introduces a “reasoning” capability that allows it to think before creating.
Details:
Ray3 brings improvements in fidelity and realism, including advances in rendering crowd scenes, light interactions and reflections, motion blur, and character consistency.
It supports up to 16-bit color depth and EXR export, making it easier to integrate with professional editing and color-grading workflows.
The model can also convert standard SDR videos into HDR.
Its most notable feature is its reasoning ability, which allows it to think in both images and language, follow complex instructions, and assess its own outputs.
Ray3 also introduces a Draft Mode that generates rough video previews up to five times faster and at lower cost, which can then be upgraded to full 4K HDR “Hi-Fi” quality in a second pass.
Luma AI positions Ray3 as a creative tool “built to tell stories,” showcasing short films created with the technology. The model is already available for free through the Dream Machine app.
AI Tutorial
How to turn sketches into images with Grok
Instead of typing long prompts and hoping for the best, Grok lets you draw your idea directly on a canvas and then transforms it into a polished image.
Go to the Grok website and sign in.
In the prompt box, click the attachment icon and select Draw a sketch.
A blank canvas will open with a toolbar. Pick the pen tool and adjust stroke size using the slider.
Draw your idea: Use your mouse or trackpad to sketch your design, just like in a basic art app.
You can also change color to white, yellow, red, green, or blue
Use undo, redo, or the trash icon to start over. When ready, click Done.
Your sketch will now appear in the prompt box. Next, describe the output by adding a prompt such as “in a cartoon style” or “as a watercolor painting,” or write something more detailed depending on what you want.
Grok will then generate image options based on your sketch. Choose one and refine it further with text prompts if needed.
AI Tools to check out
⭐ Zeabur: DevOps AI agent for vibe coders. Deploy anything to the cloud by chatting with AI.
👉 Marblism: Get an AI Team who run your inbox, socials, SEO, lead generation, calls, and support.
📧 Mocke: Mock email campaign by AI.
🦾 Tambo: An open-source AI orchestration framework for your React front end.
✍ Slashit: Turn repetitive messages, proposals, and responses into one-click templates you can use anywhere.
AI Findings/Resources
💸 A woman used ChatGPT to win the $150K Powerball jackpot
👀 New theoretical economics papers exploring the implications of AGI
🧊 Gpt-5-Codex’s attempt at one-shotting Minecraft in three.js
🔮 Where is AI taking us? | Sam Altman & Vinod Khosla
The latest in AI and Tech
Apple Watch Series 11 can now notify users of potential high blood pressure, using AI applied to existing heart sensor data.
The FDA-approved feature also works on Series 9 and later models.
Mississippi poet Talisha Jones, who creates music as Xania Monet using Suno, has signed a multimillion-dollar agreement with Hallwood Media after her songs climbed the charts.
Her track How Was I Supposed to Know hit No. 1 on R&B Digital Song Sales, while her catalog has earned nearly 10 million U.S. streams.
Satya Nadella told employees he’s “haunted” by the fate of Digital Equipment Corporation, warning that Microsoft could suffer a similar downfall if it fails to lead in AI.
His comments came as staff voiced concerns about a colder, fear-driven culture following mass layoffs.
Although Microsoft is thriving financially, Nadella warned that even its most beloved products like Office could lose relevance as AI reshapes productivity, urging a renewal of focus on building for the future rather than clinging to past successes.
Mistral has released Magistral Small 1.2 and Medium 1.2, updated versions of its reasoning LLMs. Both now feature a vision encoder for image analysis, along with performance and usability improvements.
The deal would give Meta additional computing power to train and deploy AI models, complementing its existing cloud providers.
That’s a wrap!
We'd love to hear your thoughts on today's email!Your feedback helps us improve our content |
Not subscribed yet? Sign up here and send it to a colleague or friend!
See you in our next edition!
Gina 👩🏻💻