Kimi Plan limits are just unusable right now - If you take a Kimi plan and try to use Kimi K3.1 Pro, your 5-hour limit will go very fast. I have the Max plan, and mine was gone in less than 15 minutes - If you use Kimi K3.1 Pro with High mode, the outputs are just mid, and the limits aren't good either - They should have addressed the compute issue before launch and done some deal to get more compute from xAI, Google, or others - For now, I would suggest waiting for better limits before buying their plan - And ngl, the benchmarks and real-world performance of Kimi K3.1 are different, so don't assume it's an openSource version of Opus 5.5. It's just not on that level in my experience.

I asked MiniMax M3.1 Flash to build a website where you can shatter the glass. It took 2-3 iterations, but the final output turned out pretty impressive. The model is surprisingly capable for its price, and its quite fast too. Best part? You can currently try it for free on OpenCode. Space Bunny was MiniMax M3.1 Flash all along.

Gemini 4 Carbon, 2nd checkpoint in Arena. I asked it to build a voxel-style Japanese temple valley in 3D. The output is very strong and significantly better than Gemini 4 Argon. The composition, colors, lighting, reflection and overall scene quality are a clear step up.

Gemini 4 Carbon checkpoint in Arena under the codename "gemini-4-argon." I asked it to build an interactive 3D Crystal Citrus with realistic textures, lighting, reflections, and animations. The output is good. Its 3D capabilities better than the Argon. Took around 20 minutes to build this.

Gemini 4 Carbon Leaks: Its Not Far Away - Gemini 4 Carbon EAP has started and has been given to partners for testing. Its also being tested on LMArena, and I will do my best to get access to test this model. - Its on LMArena, so people will most likely get access to it soon. - The model is an upgrade over Argon and should fix the things Argon lacks. For now, its still bad at frontend, like Argon, but better at 3D. - It could be released at the end of this month or the beginning of November. The release isn't far away. I hope Gemini 4 Carbon is much better at frontend and coding and can match Opus 5.5. Argon was a good start, but it’s time to compete.

Fable 5.5 looks close. - I tested the latest checkpoint, and its one shot performance is really good. - Partners already have access and say the newer checkpoint is stronger than the previous one. - With the IPO coming up, Anthropic has more reason to ship strong models quickly. - The strongest indication I'm hearing points to next week, most likely Tuesday or Wednesday. - We already know how good Opus 5.5 is, so imagine what Fable 5.5 could bring. Looks like we won't have to wait much longer.

Fable 5.5 one-shotted this. I just asked it to create a boat floating in a garden, and the result is beautiful. The water, trees, mountains, birds, bridge, night lights, and especially the reflections on the water. Everything looks so good, from the colors to the lighting. And it made the whole thing in one go. Fable 5.5 is looking really promising. Probably coming next week.

Fable 5.5 is coming next week. - It has shown up in Azure, as I found it in the MLflow Azure catalog, and it's also listed on Artificial Analysis under the codename "claude-onigiri-eap-2". - In Claude Code, Opus 5.5 is also routing to a new model, which is most likely Fable 5.5. - Pricing is gonna be same as Fable 5.1, as the Azure MLflow mentioned the same pricing. - It's also in internal testing, and with all these signs launch not too far, They're also targeting this month for the release. So, there's a high chance Fable 5.5 launches next week.

Fable 5.5 showed up in MLflow’s Azure catalog. - Its the 5.1 row with a new name. Pricing: $10 / $50, cache read $0.25. The only thing that changed is the date, Oct 14. - Deprecation is Dec 5, 2027, the same date already listed for Fable 5 and 5.1 in that file. Fable 5 cache read there is still $1.00. - Pricing is the same as Fable 5.1, and its visible in multiple places, like Azure and Artificial Analysis, under the codename "claude-onigiri-eap-2". All of these signs make an immediate launch likely this week or next week.

Guys, officially I don't have early access to Gemini 4 Argon, but through a friend who has access I managed to get early access (tbh I wanted it directly from @GoogleDeepMind next time consider me on the list!). Here is the second test I created with Argon: I asked Argon to create an Impossible Museum. Tbh I am loving this model. Its speed + creativity in 3D is on another level. Full review coming soon.

Opus 5.5 My Impressions / Review - Its my daily driver model right now. For the price, its one of the best models available. - Its quite fast, better than GPT-6.1 Sol, and in many areas I would even take it over Astra. - Its especially good at motion design, sound design, and also pretty good at basic video editing. - Very good at 3D web design, physics, backend and databases. - Its also really good at long-horizon tasks. I can give it tasks that run for 7–8 hours, and it keeps working well. Subagents are also quite good, although they can burn through a lot of tokens pretty easily. - All of this makes Opus 5.5 my choice for a daily driver, mainly because of the combination of speed, quality and cost. What model are you guys using as your daily driver? And if you have used Opus 5.5, what’s your take on it?

Made this interactive 3D honeybee using GPT-6.1 Sol Took 3-4 iterations to refine the fur, eyes, translucent wings, and lighting for a more realistic feel. Built with Three.js and WebGL, with camera controls, wing animations, and a short walk cycle.

GMI Cloud just raised $668M in new financing, with a Series B led by ARCHIV and NVIDIA participating, plus a credit facility led by CTBC. What I find interesting is what this infrastructure enables: startups can get GPU capacity for training and inference without having to lock up huge amounts of compute years in advance. As AI apps move toward agents and millions of users, reliable compute becomes just as important as the model itself. They are already seeing 9×+ growth in contracted ARR and 4T tokens processed every week. Curious to see what they are cooking next. @gmi_cloud @alex_yehya

Asked GPT-6.1 Sol to create an interactive 3D Glass Desert Pavilion. Sunset lighting, glass walls, water reflections, and cinematic camera views. Its quite impressive and feels far better than GPT-6 Sol. The cost to quality is really good. I still had to do 3-4 iterations to get it close to perfect, but the output is seriously impressive.

Opus 5 is really amazing at motion design. I gave it one prompt: make a 30s film where the camera keeps zooming into letters, and every letter reveals a new word. 18 words, all hidden inside each other. The typography, animation, camera moves, music, beat sync, it made everything itself. One prompt. Around 50 minutes.

Opus 5.5 is insanely good at motion design. I gave it one prompt: "create a motion design video of a poster breaking out of its own frame." The animation, the camera moves, the sound design, even the music. It composed the whole soundtrack itself and synced every hit to the beat. Sound on 🔊

Tested a stealth model called Space Bunny with one prompt: turn a resume into a retro game. Fed it a fake test profile, and it built a playable career map with 3 unlockable levels, player stats, and a final-boss skill tree. Single HTML file. Free on OpenCode this week

Claude Opus 5.5 built a 300 metre tall flower that blooms in real time. Day-to-night skies, dynamic rain, fireflies, and even a cutaway view. Zero assets. Just code. Cost: 10M tokens.