Jan Shi

@hector_podcast @MrAhmadAwais Yea I keep seeing claims how other models like minimax 2.7 or Kimi 2.6 or GLM 5.1 is significantly better and how v4-flash has too much hallucination rates. Idk maybe I haven’t tested each out enough just rotating around in Hermes to see the best performance/price right now

@MrAhmadAwais One of the only things keeping my Hermes agent alive

how the heck is DeepSeek v4 flash so good and so cheap?

@pa3ko_phurn @AntiGlobalisttt @Paddvart @Kendo_Arsenal @FabrizioRomano Where is his hand supposed to be?? Above his head??

@AntiGlobalisttt @Paddvart @Kendo_Arsenal @FabrizioRomano No na leg ball Can’t u see the ball is between his hand and his stomach

🚨 Kompany: “Why is it not a red card? I don't understand. Why we concede a penalty in Paris and we don't receive one here? I understand the rules, but the hand was above his head”. “It felt like there was a hand that tilted it to the wrong side for us”.

@affprinter Do you just delete the account or the whole app if it’s stuck at ~200 views after 3 posts

i was the guy posting 5 TikTok slideshows a day on a fresh acct wondering why nothing worked every post died at 200 views and i thought it was content it was never content. i was failing trust scoring before tiktok even saw a slideshow full breakdown of the 72 hr fix in the article

@basedautismo @levelsio I use these. Pain to clean. But I sousvide just a handful of times a year. Only needed for duck and thick bone-in steaks

@levelsio With proper silicone bags it's fine amazon.com/gp/aw/d/B07GCV…

Sous vide is a microplastics nightmare

@abdullah_rang @championswimmer M2.7 is a definitely tier below k2.6. It’s closer to DeepSeek v4-flash than k2.6

@championswimmer Have you given minimax2.7-high speed a shot? Pretty similar results based on what you mentioned

Kimi K2.6 makes Sonnet and Haiku totally dead in the water. It can do 85-90% of the tasks I entrust Opus with as well. Connect some search MCPs (Parallel/Exa/Perplexity) and it beats Opus at way less cost, even including the search API costs. Totally stopped using Opus now.

@aiCrof I've been curious about Crof for a while! Idk if its better value than Opencode bc plans based on the number of api requests can burn through quickly with tool calls without consuming lots of tokens.

@CookingRobotGuy Let us know if you wish to try Crof out!

Crof.ai users, What do you usually use Crof for? Long coding tasks, Openclaw or just as a chatbot? Let us know below 👇

prompting >> which frontier model you choose. Spent 100million+ tokens debugging this issue, feeding GPT5.5 + Opus 4.7 logs, spinning up Kimi 2.6 to implement tests. Nothing worked until GLM5.1 pulled up, reviewed the logs, came up with a plan that GPT 5.5 tweaked. Boom fixed!

@reach_vb If anyone wants to give it a try, shoot me a DM with your Apple App Store test flight email

@reach_vb Built an iOS ssh terminal so I can command my Mac/linux and use codex cli while traveling! Smoother, more reliable, and stable than terminus imo!! Persistent terminal sessions—survives network disconnects and even you killing the iOS app and cold launching.

Weekend hack: Build with GPT-5.5 + Codex. Drop your demo in the replies. #1 by likes: 1 year of ChatGPT Pro 2 runner-ups: 6 months each Bonus: Codex picks a wild card winner. Enjoy!

@affprinter What do to when you’re stuck on 200views on your first post? Delete the app and start a new account?

I cracked the TikTok account warmup algorithm heres exactly what tiktok is measuring in ur first 72 hours and how to pass every check: tiktok doesnt trust new accounts. theyve been burned by bots too many times. so they run every new account through a trust scoring system. most ppl fail it without knowing it exists. heres what triggers a failed trust score: posting within 24 hours of account creation adding links to bio on day 1 following/unfollowing in rapid bursts only engaging w content in one narrow category posting multiple times before getting any engagement using images that match their duplicate database heres the warmup protocol that gets me 10k+ views on first posts consistently: DAY 1: scroll FYP for 30+ min like 15-20 videos in ur niche leave 5-7 genuine comments follow 8-10 accounts that post content like urs DO NOT post anything DO NOT add any links DAY 2: same engagement pattern search for keywords related to ur niche watch 10+ videos to completion in that category let the algorithm learn what ur about still no posting still no links DAY 3: post ONE warmup slideshow not promotional. just engaging content in ur niche. use a trending sound from last 7 days images must be unique (ill explain how later) watch for 24 hrs if that warmup post hits 1k+ views ur cleared. trust score passed. now u can post promotional content. if it gets under 200 views ur account is cooked. the algorithm flagged u. start over. this process takes 3 days. most ppl wont do it bc theyre impatient. thats exactly why most ppls accounts get shadowbanned permanently. the warmup isnt optional. its the foundation everything else is built on. skip it and nothing else in ur strategy matters.

@Jason @WisprFlow @elgato Is wisprflow better than Handy, which runs locally on your Mac/PC?

Get @WisprFlow and the @elgato foot pedal — it’s life changing

@NousResearch @dontlockmyacc0 @ComfyUI Dude even hermes doesn't know everything about hermes!! Using k2.6 and asked to connect my gmail. and it wanted me to go through himalaya or google api instead of native email gateway.

@dontlockmyacc0 @ComfyUI Just ask Hermes about Hermes!

ComfyUI is the most flexible, composable, and powerful open-source media generation tool with a massive ecosystem of workflows and custom nodes. Your Hermes Agent can now install, launch, manage, and run sophisticated @ComfyUI workflows on demand.

Bro fix this. /status and statusline limits are always off @OpenAI

@theio666 @FactoryAI Yea I used minimax m2.7 for weeks as well. It’s fine at tool calling and stuff but hallucinates too much, esp on the first attempt. If I ask Claude or codex to review m2.7’s analysis/conclusions, it’s usually full of fallacies and lazy logic. K2.6 definitely performing better

@CookingRobotGuy @FactoryAI 1) M2.7 is a big upgrade over M2.5 2) it's great for agentic tasks, I don't make it write code for me - for that I have codex (on my main codebase even opus isn't reliable), but review-verify loop, where I use pipeline like on the screen - M2.7 does it job quite well.

Which model reviews code best? We benchmarked 13 models on AI code review across real PRs and the results are surprising. Spending more tokens did not result in better code review. A $1.25/PR model beat another that was more than 2x the cost. Meanwhile, budget models at $0.15/PR delivered ~80% of the quality of frontier models while being 10-30x cheaper. In fact, cost only explained ~21% of the difference in code review quality.

@theio666 @FactoryAI Yo minimax 2.7 is NOT reliable for code . At a minimum you MUST ask "are you certain? did you trace through all the code?" to get even 60-80% accuracy. but of course, depends on how complicated the code is. But that's my experience since minimax 2.5 days. K2.6 is much better.

@FactoryAI As someone who's playing with Minimax 2.7 for reviews lately, it's a bit tricky to make it work. In my review pipeline it's performing quite well, but it's important that I allow it to execute code and create/run new tests, not just read the diffs. Changes final quality a lot.

@PavitraGolchha @mhdcode @anomalyco its hella slow during prime time. I'm already 50% through my tokens for the month and 100% of tokens for the week after 3 days of ripping k2.6. Don't even bother with GLM5.1, you get like 2 long sessions with it and poof it's gone.

@mhdcode @anomalyco I should probably cancel Copilot sub and move to OpenCode Go

everyone at @anomalyco is skipping lunch to give you this for $10

@Suyash151504 @0xSero Created a more reliable version of iOS remote terminal. Better than terminus imo esp when you disconnect or background the app. Working on the UI now. LMK if you want a free test flight for it

@0xSero Are you running this in phone, it seems like a phone layout to me ? 🤔

I can’t believe I’m saying this but Opus-4.7 in Droid is worth the money. Why can’t it be this good via Anthropic’s subscriptions );

@0xSero @badlogicgames Do you find it reliable? Does it drop connections/packets on the go?

@badlogicgames termius.com <-- the app runs pi too.

@Kumohira1 @LottoLabs @iso20xl It’s good but I’m already at 25% monthly usage and 51% weekly after one day.

@LottoLabs So, if I grab the plan today, I can run it (vibe code) without hitting monthly limits? @iso20xl @LottoLabs

Opencode go and Kimi k2.6 are running 24/7