Latest Update

Claude Fable 5: Anthropic's most powerful public model

2026.06.10
What's up, Carat crew here. 😊
Today: Anthropic's most powerful public model Claude Fable 5, Luma Ray3.2 that turns direction into cinema, Google's instant live translation, and Apple's 20-billion-parameter model running on your iPhone.

🔥 Anthropic's most powerful model, Claude Fable 5, is here

ⓒ @EHuanglu
Anthropic released Claude Fable 5. It's the most capable model anyone can use, and it's actually the same model as Mythos 5, the top-tier model that used to go only to security agencies. Anthropic just locked a few risky capabilities and opened it up to everyone.
It's state-of-the-art at coding, knowledge work, and vision, and the longer and more complex the task, the bigger its lead over other models. In risky areas like cybersecurity, biology, and chemistry, though, it blocks the answer and falls back to Claude Opus 4.8.
Right after launch, a demo of it beating Pokemon FireRed from screenshots alone went viral. No map, no coordinates, no game memory, it played through using only what's on screen.

📌 3 more things today

1️⃣ Translation while you're still talking? Google Live Translate

ⓒ Google
Google released a real-time speech translation model called Gemini 3.5 Live Translate. It starts translating the moment you start talking, so it doesn't wait for you to finish your sentence.
It supports over 70 languages and keeps translating your earlier words while you move on to the next ones. No awkward pauses or choppy gaps. You can use it in Google Translate, Meet, and AI Studio.

2️⃣ Direction goes in, cinema comes out: Luma Ray3.2

ⓒ Luma
Luma released a new video model, Ray3.2. Its pitch is 'direction goes in, cinema comes out,' and you can now direct scene continuity, camera moves, and mood much more precisely.
Instead of making one cut at a time, it keeps the tone and flow consistent and renders a cinematic video in one go. You can make this kind of cinematic video with Luma's video models on Carat too.
👉 Make it on Carat

3️⃣ Apple is putting a 20-billion-parameter AI inside the iPhone

ⓒ Apple
ⓒ Apple
Apple unveiled an on-device model called AFM 3 Core Advanced that runs on the iPhone without internet. It has 20 billion parameters, but it only switches on 1 to 4 billion at a time, so it can run on a phone.
The full model sits in storage, and it pulls in only the parts it needs for each request. It improves voice and dictation accuracy, and it's available on the iPhone 17 Pro and high-end Mac and iPad models.

🧪 Prompt of the day

Turning your photo into a MapleStory character is all over Instagram right now. Drop in one photo and it becomes a Maple-style 2D character. The Carat team tried it ourselves. Left is the original, right is the result.
Left: original → Right: generated with ChatGPT Image 2 Medium on Carat
Left: original → Right: generated with ChatGPT Image 2 Medium on Carat
Transform my photo with the following prompt using ChatGPT Image 2 Medium.
Prompt: Transform the uploaded image into a MapleStory-style 2D chibi game character illustration.

This is a universal image-to-image character conversion prompt: whatever person, avatar, anime character, mascot, or fictional character appears in the uploaded image should be converted into an original MapleStory-inspired SD game avatar while preserving the key recognizable traits from the input image. Keep the subject count, main pose, silhouette, hairstyle impression, facial expression, outfit category, color relationships, and major accessories from the uploaded image, but redraw everything in a cute side-scrolling MMORPG character-art style.

Style direction: MapleStory-like 2D online RPG avatar art, super-deformed cute proportions, oversized head, compact tiny body, large expressive eyes, simplified nose and mouth, clean thick outlines, crisp cel shading, glossy hair highlights, bright polished colors, soft gradient accents, sprite-like compact shape language, playful Korean online game aesthetic, high-resolution 2D illustration.

Character conversion rules:
- Preserve the uploaded image's identity cues and composition as much as possible.
- If there is one subject, make one Maple-style avatar.
- If there are multiple subjects, convert each subject into a matching Maple-style avatar while keeping their relative positions.
- Translate real clothing or complex costume details into simplified cute game-avatar clothing.
- Translate props, weapons, hats, hair ornaments, bags, or accessories into simplified fantasy game item details.
- Keep the result original and game-avatar-like, not a direct copy of any existing franchise character.

Composition: centered character portrait or full-body avatar depending on the uploaded image, clean simple background, polished character selection icon feeling. The final image must clearly look like a Maple-style 2D character conversion, not a realistic portrait, not a generic anime illustration, and not a 3D render.

No text, no captions, no watermark, no logo, no photorealism, no 3D render, no realistic skin texture, no complex painterly realism.
Make it yourself on Carat right now. 👉 Make it on Carat
That's it for today. We'll be back tomorrow with fresh AI news. ☺️

Was this helpful? Share with your friends

Share

Get daily AI trend updates