Saadi (Digital)

Everyone is basically training these models. People pay for subscriptions, but the reality is that AI companies need users and their data more than users need the models.

@CommandCodeAI @hogopaiva yeah we understand , great work it's a SOTA . GLM 5.2 should be everywhere

@hogopaiva it's an expensive model. #models" target="_blank" rel="nofollow noopener">commandcode.ai/docs/resources…

GLM-5.2 Fast is now live on Command Code! The high-throughput build of GLM-5.2, same frontier coding, now at 120–250 tokens/sec for snappier long-horizon runs. 1M context. Tool use. Reasoning. Available in the $1 Go plan with $10 usage credits and above!! Try it out now with /model.

@zyloo_io Same JSON shape as OpenAI /v1/chat/completions

@hqmank they are tight on compute right now . hitting limits faster , and they are allowing less image gen than before .

Codex Pro subscribers, check your 5-hour usage. Multiple users are reporting unusually fast quota drain today. Some suspect their Pro accounts are getting Plus-level limits, followed by: "Selected model is at capacity." Are you running into this too?

@ore825 getzaki.com available on play store and app store . major update coming with a lot of recipes and ingredients . go.getzaki.com/zaki-ios go.getzaki.com/zaki-android

この人天才だろ。じゃがアリゴ開発した時思い付かなかった、悔しい

@suzuka_fz1124 getzaki.com available on play store and app store . major update coming with a lot of recipes and ingredients . go.getzaki.com/zaki-ios go.getzaki.com/zaki-android

味仙食べてがんばるぞー！ もうすぐ会えるね♡♡

@elonmusk yet u are the most racist billionaire. can u bear criticism and what does your common sense say ? the GrokAI is still behind even after acquiring Cursor.

Accusations of racism have been weaponized to stop criticism and common sense. No more.

@AZK_SA well if u don't know coding , the LLMs providers will make a monoply , no option scenario for the world

مؤسس تليجرام، ينصح الطلاب: لا تضيعوا وقتكم في تعلم البرمجة والذكاء الاصطناعي.. أدرسوا الرياضيات و الفيزياء جيدا. و إيلون ماسك يتفق معه في رد على تغريدته 👍

@RayFernando1337 @giuseppe_1337 is it better in droid , i think in claude code it's the best . i left using droid 6 months ago , is it good with glm ?

@giuseppe_1337 I’m using it in droid and if I have spec mode turned on it ships features.

GLM 5.2 tokens per second HOLY MOLY Baseten! It cost me about $1 to ship a small iOS feature using the droid harness.

@hqmank i have used it in Z Code also . but in claude code it's so much better . even mimo 2.5 pro is great in Claude Code . atleast better than Sonnet 4.6 High.

When ChatGPT and Codex go down, what's your backup? I'm using Z.ai GLM-5.2 in Claude Code.

@bridgemindai nope , just US is holding back . they are waiting for the demand to rise . the Asset Management companies are planning how and when to launch .

The U.S. government is destroying AI right now. Fable 5 has been completely suspended (no updates). GPT 5.6 has been delayed (twice now). AND Gemini 3.5 Pro has been pushed back. This isn't just about Fable 5 anymore. It's about all frontier models.

@sadik_0x @Dispropoganda yeah dollar seems fine too . but i want to keep it riyal

@codersaadi @Dispropoganda Dollar not riyal

@bridgemindai 200$ loop is the worst . even with the best plan . honestly i don't see much improvements in Opus . 4.6 4.7 4.8 to me there is just a 5% difference in all of them .

I just cancelled two of my three Claude Max subscriptions. Claude Opus 4.8 is not worth $600/month. Fable 5 was. Fable 5 has been offline for weeks. Anthropic's communication has been terrible. No restoration timeline. No signal on what comes next. I am running GPT 5.5 xHigh for backend, Fable 5's frontend crown is sitting empty, and GLM 5.2 is closing the gap from below. The vibe coding movement is moving on. It has to.

@MartinSzerment @hqmank yeah other than that i like how no vendor lock , like if i need Opus 4.8 only claude can give me the best value . that's a vendor lock but with GLM ,Mimo , Kimi, Deepseek . it doesn't matter maybe some other subscription is giving more compute and value than their own providers

Exactly, and that is the whole point. Open weights are the only real exit, because you cannot ask a passport from a model running on your own GPU. Every ID gate Anthropic adds is free advertising for whoever ships weights with no checks. Is it a moat, or a recruitment poster for the competition?

Reports suggest Anthropic may use Apple Wallet Digital ID for Claude identity verification. If Fable 5 returns for verified U.S. users only, the path is obvious: passport in Wallet → Claude verified → model unlocked The next AI gate may not be a paywall. It may be your ID.

@hqmank @MartinSzerment even china is already doing it , see how BigModel is china only (ZhipuAI China) , compute and cost is completely different . also other chinese companies are doing the same . the more people use it , more feedback they give and the better it becomes.

@MartinSzerment Very likely. That’s why I hope open-source models keep catching up.

@MelodicHouseMix @hqmank i'll examine it once i get more time . that write limit thing is real you will see a red light blinking near your power button . i thought the issue is related to tool calling in antigravity or some MCP

@codersaadi @hqmank antigravity'de buna benzer bir sorun yok.

1/ Codex is quietly killing your SSD. It writes diagnostic logs to disk non-stop, even when you're not doing anything. Your SSD has a write limit. Codex is burning through it in the background. One command fixes it 👇

i think google should learn something from GLM and Kimi . both Moonshot and zhipu ai are doing a great work (i don't disregard Deepseek it's great fast and very valuable if used well)

@DanielLockyer why use it ?

I hate it when third-party libraries have some sort of built-in caching, but have zero reference to it in the README and no way to turn it off Not only does it break the mental model of what the code is doing, but it unnecessarily bloats memory where it isn't wanted

@yacineMTB nice group photo of claude pro users .

strapping in your toddler to ultra a/b tested mega optimized hyper stimuli and telling other parents that it's okay, because the one person (out of the team of 50) has a "behavioral psychology degree" while she hypnotically plays sounds and switches scenes every 5 seconds