CPM

@BenjaminUIX Wow.. what do you use to animate it?

Animated version just brought it alive! 🔥

@nexxeln @opencode Is there a getting started guide?

since everyone already knows, i’m joining @opencode here's a demo of a thing im working on: async subagents / background agents should this exist? tell me what you’d use it for

@bananababydoll Love your work, how to do you ensure face consistency across images?

can’t sleep again. Nano Banana Pro {   "meta": {     "quality": "ultra photorealistic",     "resolution": "8k",     "camera": "iPhone 15 Pro Max",     "lens": "24mm",     "aspect_ratio": "9:16",     "style": "low light iphone selfie, soft grain, intimate bedroom realism"   },   "character_lock": {     "age": "mid 20s",     "ethnicity": "Mediterranean",     "hair": { "color": "dark brown", "style": "messy down" },     "body": { "type": "curvy", "chest": "full", "waist": "tiny", "hips": "round" }   },   "scene": {     "location": "bedroom",     "time": "late night",     "atmosphere": "quiet, needy, intimate"   },   "camera_perspective": {     "pov": "we ARE her phone",     "angle": "slightly high",     "framing": "face + chest + stomach"   },   "subject": {     "position": "lying on bed",     "pose": {       "one_arm": "behind head",       "expression": "sleepy eyes, lips parted"     },     "outfit": {       "top": {         "type": "tiny spaghetti strap tank",         "color": "white",         "fit": "tight, braless"       },       "bottom": {         "type": "cotton panties",         "color": "light grey"       }     }   } }

@ivanfioravanti What’s the difference between using say kimi with CC vs OC ?

@Cuan It's pretty solid, I alternate it with OpenCode.

Kimi K2.5 (Kimi CLI) vs MiniMax 2.1 (CC) vs GLM 4.7 (CC). 🔥 Same prompt to create a single-page website for "PHANTOM PROTOCOL" a fictional tactical shooter video game, 0-shot. Spoiler IMO: 🥇 Kimi K2.5 is another league 🥈 MiniMax 2.1 🥉 GLM 4.7

@ivanfioravanti I did not know it could use other models ..

@Cuan Here a quick how-to. x.com/ivanfioravanti…

@ivanfioravanti Rock star

@Jarodxu7 Happycapy

Excited to launch happycapy, an agent-native computer on your browser. Claude Code → Clawdbot → Happycapy > Private sandbox for everyone to run Claude Code anytime anywhere - no Mac Mini needed > GUI built for everyday user, visualizing skills and its output

@alexocheema @exolabs @awnihannun Is there any starter guide ?

Running GLM-4.7-Flash on 4 x M4 Pro Mac Minis using @exolabs. Uses tensor parallelism with RDMA over Thunderbolt & MLX backend (h/t @awnihannun). Runs at 100 tok/sec. We're working on optimizing this at @exolabs. Aiming to hit ~200 tok/sec on this setup soon.

@Mike_Scully_ Auto

I literally never run out of content anymore. I built an automation that watches 50+ news sources, trends, viral posts, filters out the noise, scores what actually matters and turns it into ready-to-post content. It spots trends in my niche before they hit everyone’s timeline. No scrolling. No guessing. No “what should I post today?” Saves me easily 20+ hours a month and keeps me ahead of trends. Comment “Auto” and I’ll DM it to you. (Follow me so I can send DM)

@michelleefang Wow very generous

if you're vibe coding or building over the holidays, i want to gift one of you a 6 month subscription of claude pro to support <3 just drop a comment below. merry christmas!

@benln Building an agentic vulnerability management solution ..

Who's building over the holidays? Will add you to a group chat I started on X.

@lando_garrison Do you not find them very very slow?

GLM 4.7 and MiniMax 2.1 are the first time I genuinely feel Chinese made models have surpassed American models. Both models may not be better in every way, but they feel infinitely easier to use and get what you want, plus cheaper (this is from my experience)

Do not upgrade to Gemini 3 (pro or flash) if you are using tools. There are many severe bugs such as outputting JSON in the wrong place, or ignoring tool output.

@maccaw We have the same issue.. we think it’s related to the reasoning signature getting lost

@gdaybrice @aisdk We’re having the same issue.. we think it’s related to the reasoning signature getting lost after a tool call response

getting weird answers from gemini 3 pro and flash today with @aisdk. It outputs text as json with {"thoughts": "...", "response": "..."} completely fine with 2.5 pro

@GoogleAIStudio Hoping you can fix the issue with it using tools.. it messes up the json response and errors out

⚡️ gemini 3 flash is here • frontier intelligence built for speed • pro-grade reasoning at a fraction of the cost • our most impressive model for agentic workflows available now as gemini-3-flash-preview in AI Studio and the Gemini API

@OfficialLoganK It is unable to work with tools at all.. while it calls them, the Json response format is all over the place and it errors out…..@officallogank

Introducing Gemini 3 Flash, our frontier intelligence model, available at scale for everyone. It excels at coding, tool calling, and is stronger than 2.5 Pro across most metrics!! ⚡️ Available in the API at $0.50 in / 1M tokens and $3.00 out / 1M tokens across.

@maccaw Thanks .. we were hitting the same issue,.. no obv workaround yet

More info here: discuss.ai.google.dev/t/gemini-respo…

@Zephyr_hg Gamma

My sales email response rate jumped from 5% to 18%. Built an automation that creates personalized sales presentations in 90 seconds by researching prospect websites and uses Gamma to generate custom pitch decks automatically. Scrapes their sitemap, analyzes key pages, identifies pain points, and builds an 8-slide deck tailored specifically to their business. Prospects can tell I actually researched them because the deck references their specific challenges. No more generic templates that scream copy-paste. Comment "GAMMA" and I'll DM it to you (must be following)