Atri

Building the next generation of AI agents, but slowed down by the crippling AI psychosis

[Atri @CognitiveTake]

@nahcrof I see you have q8 have you quantized both weights and activations? Or just the weight?

Would be good if you can clarify what have you done to reduce the price.

[Atri @CognitiveTake]

@adamwathan @AmpCode People need to separate the business of harness from token reselling. Absolutely no reason for amp to try and resell tokens.

[Atri @CognitiveTake]

@ddalcu @ddalcu not able to use it in m4max 128G. It says “MISSING WEIGHT: model.embed_tokens.weight”

Not sure what am I missing.

[Atri @CognitiveTake]

@opencode GLM 5.3 flash is not even on the chart. People liked ox alpha because it was free. The limits for it now are abysmal. It was all a marketing gimmick. Back to Deepseek flash we go. Really shows who is the real GOAT.

[Atri @CognitiveTake]

@elshayib_ This is all fake. I have the 60 dollar plan. 4% monthly usage on one task that lasted an hour. Grok 4.6 high(not fast). I calculated 400 million tokens per week at this rate. This is ok, but nowhere near what these people bought by musk try to claim. Still lower than codex

[Atri @CognitiveTake]

@NousResearch @cyrilXBT @bot The IQ of the average X “influencer” is less than a 10 year old. What can we even expect. They don’t even understand the basics of a product before talking about it.

[Atri @CognitiveTake]

@antirez I just hope you don’t leave us metal users stranded man. Very few people are even thinking about us.

[Atri @CognitiveTake]

The person I trust blindly on X. Holding the fort for local inference on apple ecosystem

[antirez @antirez]

Btw the Redis story repeats itself: I'm working at DwarfStar for free for the community and because I enjoy it. But I'm receiving criticisms, since people are worried that this will break their AI-richness plans. I want to say to everybody thinking that I should stop that each time you tell me this, I'll double down my efforts towards a completely no profit engine for local inference. Better to shut up basically.

[Atri @CognitiveTake]

@antirez Massive respect for you. You are easily the person I trust the most on X.

[Atri @CognitiveTake]

@r1othecreator @uzairakrum Muse spark is free. Only a fool would use it on subscription.

[Atri @CognitiveTake]

@Abdi_Akhmet @uzairakrum How is it peak? They give 15 dollars for 10 dollars and then enforce 5 hr, weekly and monthly limit. At this point I would just pay api prices. 5 dollars extra is a joke. Not a single model you would want to use today has 60 dollar limit anymore.

[Atri @CognitiveTake]

@SourceCodeplz @uzairakrum They are not giving it on any new models. Any model you would want to use right now doesn’t have it. Even the Deepseek flash vision

[Atri @CognitiveTake]

@jayair Let’s see how many tokens it does today with your horrendous limits.

[Atri @CognitiveTake]

@BearHuddleston @MiaAI_lab The recipes are up from actual reliable sources like vllm and sglang. Why would you want the recipes from fake influencers like these

[Atri @CognitiveTake]

@FishRaposo @Father_Russia69 @Da7_Tech Arrogant and a prick. False advertising. Trapping people by promising something else and then delivering something else. Blocking people who criticize them instead of addressing it.

[Atri @CognitiveTake]

@Father_Russia69 @Da7_Tech lol. We don’t talk about scamsters here.

[Atri @CognitiveTake]

@ivanfioravanti I think Qwen is the bigger release for me . That intelligence in such small size is impressive. GLM is fine. Deepseek flash vision is already close, one more round of RL and it would be there probably. And DS is still cheaper(cache) for roughly same model size

[Atri @CognitiveTake]

@Zai_org Running on Chinese chips and yet this expensive for this model size? I was hoping you would give original Deepseek flash level pricing, but you have made it more expensive as compared to the current Deepseek flash pricing( Because of cache pricing)

[Atri @CognitiveTake]

@thdxr Why is every model now giving only 15 dollar allowance? Even for small models like this you give 15 dollar monthly allowance. Makes no sense to continue using opencode go anymore

[Atri @CognitiveTake]

This is what they were hyping? Look at the usage of GLM flash. This is pathetic.