AGI Pulse

@RomeoOnyi @burkov Since I saw that fact here first laying on my bed, I thought why not confirm rn instead of picking my lazy ass up and turn on every electric device I need to before I try inserting a picture on zcode

@AGI_Pulse @burkov Why not try by feeding it a picture??

For the last three days, I've been using GLM 5.2 with OpenCode instead of Codex and I don't see any difference. There wasn't any bug that GLM would fail to fix or a feature it would fail to add as requested. The only downside is that this model cannot see, so if it's simpler to explain an issue by pasting a screenshot, I would still use Codex. Otherwise, GLM would be my choice. Will continue to use it for two more weeks and, if it keeps just working, I will cancel my $100/month subscription with OpenAI. I already cancelled my Anthropic subscription and have no regrets. No moat isn't hypothetical anymore.

Why is my timeline flooded with a positive set of GLM 5.2 tweets?

@ai_for_success I won't place my bet on it. Specifically after some of the researchers there resigned and joined rival AI camps. Only Google can now save google.

There is a lot riding on Gemini 3.5 Pro. Beating other models on benchmarks won't be enough. Google DeepMind needs a model that makes people genuinely feel they're back in the race.

@rileybrown Prepare to pay a fortune for a bunch of Nvidia Blackwell GPUs. Don't cry river when you can't use these "super hardwares" to run high end models that come after a year or two.

If I had a budget of $10,000 to build a local setup to run GLM 5.2 locally AND any future local model that comes out soon what should my computer set up be?

@bindureddy If my institution decides to buy these crazy subscriptions, I'd simply ask them to get a subscription for cloud hosting GLM 5.2 and DS v4. Paying exorbitantly for closed source models does not make annnnnnnnyy sense to me. None at all.

Haha - this is unbelievable! I suspect you just need Opus 4.8 and GPT 5.6 for just 10-15% of tasks Kimi and GLM will do the rest Enterprises who are paying eye-watering Anthropic bills are officially OVER PAYING like crazy 🤯

@ai_for_success It gets my vote. Been fantastic till now. Been only using it through ZCode daily credits.

GLM 5.2 couldn't have arrived at a better time. While the U.S. government was busy banning Fable 5, GLM 5.2 quietly launched and seems to be winning over the community. I've not seen a single negative post about it so far. Open-source FTW 🤟

@mark_k No. Mythos model was ahead 7 months. No other models are going to come near it unless AI starts making AI.

Do you expect GPT-5.6 to beat Claude Fable 5?

@ai_for_success Maybe they are telling a story: Gemini's future isn't so bright.

🚨 John Jumper, Nobel Prize winner is leaving Google DeepMind after 9 years to join Anthropic. Yesterday it was Noam Shazeer joining OpenAI. Today it's John Jumper joining Anthropic. What's happening 👀

@BrenBuilds @jietang @elonmusk @teortaxesTex 100%. Running unquanticized GLM 5.2 would probably require 10-15 Blackwell GPUs. Better to get a subscription to any website hosting it.

@AGI_Pulse @jietang @elonmusk @teortaxesTex It's a trap 🪤

I think GLM 5.2 points to a 7 months gap currently It's around Opus 4.7-4.8 level, all told (modulo vision which in Opus's case is garbage anyway). Mythos reached Preview status (≥ Opus 4.8, functionally) by early Feb 2026. This means full PRC Mythos ("Fable") by Nov-Dec'26.

@FredaDuan used it already. It is one of the best for development purposes. You must use it on a host that has fast output otherwise it feels so slow on Z.AI itself and its IDE

I’ve been surveying builders on GLM 5.2’s actual performance - "don’t trust the benchmark, listen to what builders actually think." Would love to hear feedback from anyone with hands-on experience using GLM 5.2 or other open-source models. Quality Feedback has been consistently positive. Many say GLM 5.2 is roughly at Codex 5.5 medium-thinking level for coding. The common view is that it is stronger on front-end tasks than back-end tasks. Main pushbacks I’ve heard: - Still lags on complex problem-solving. - Weaker generalization. Price Effective cost seems to be around 20-35% of $Opus 4.8, depending on workload. Cheaper, but not as dramatic as the 4-6x gap implied by headline token pricing. How to think about open source Gavin Baker framed it well: “Frontier captures 90% of value; open source carries 80% of tokens.” But that really depends on the quality gap between the two groups. Deepseek as the cost floor $DeepSeek is by far the real cost leader. It creates a price-performance floor: any model more expensive than $DeepSeek needs to be materially better, or users may just use $DeepSeek. The pushback I've heard is that $DeepSeek can be unstable. Builders complain about inconsistent quality, higher retry rates, and more cached tokens, which can reduce the apparent cost advantage. Pricing nuance GLM may look 4-6x cheaper than Claude Opus on list pricing, but actual cost depends on: - Cache hit rate, especially in cache-heavy agent loops. - Success rate / retry rate: If one model finishes the task once and another needs 2-3 retries, token savings can disappear quickly. - Stability. Assuming 95% cached, 4% fresh, 1% output, $Opus costs ~$0.925 per 1M total tokens, while $GLM-5.2 costs ~$0.347. $GLM is ~2.67x cheaper. Assuming 90% cached, 5% fresh, 5% output, $Opus costs ~$1.950 per 1M total tokens, while $GLM-5.2 costs ~$0.524. $GLM is ~3.72x cheaper. So the more cache-heavy the agent loop, the less dramatic GLM’s cost advantage looks versus headline list pricing.

@BrenBuilds @jietang @elonmusk @teortaxesTex Don't even ask my bro

@jietang @elonmusk @teortaxesTex what GPUs will i need to run it. tell me now so can get ahead of the curve.

Not as fast as Codex, but who cares about speed when you have an open source model as good as Opus 4.8?

@yamedoff Even better

this is just codex lmao

@rileybrown Who cares

We already knew the Chinese companies relentlessly copy the US models... But now they are creating IDENTICAL apps that run those models. This is an exact replica of Codex.

@ChrissGPT So basically Anthropic is saying they are capitulating

The NY Post is reporting that Anthropic has floated a proposal to Commerce Secretary Howard Lutnick to end the US ban on its Mythos and Fable models!! The proposal includes working more closely with the White House, improving communication with the administration, and moving faster to resolve security concerns around the models. The report says Tom Brown and Sarah Heck are taking leading roles in the talks, with officials still concerned that Anthropic treated the safety issue as too narrow. I remember the earlier talk centered around jailbreaks and anthropic basically saying that it’s an inexact science, so I’m not too sure how the White House is going to respond because it doesn’t seem like they’re making jailbreaking concessions.

@ai_for_success Insiders should leak the model

Who will save Anthropic ?

@uzairansar Good catch. It simply means 3.5 Pro is simply not good enough to be released to the public.

Did Google hope we forget about 3.5 Pro? 😂 Tomorrow marks the 1 month mark since they said "3.5 Pro next month"