@shownotover Bro.... GPT 6 astra is $50 per million output tokens and $10 per million input tokens, you are not supposed to use it on $20 plan for any work anyways
@JustLingonberry I think models are already very good at coding in my opinion, they have one big issue though with hallucinations, recalling correct things, and knowing when to research the web on topics it may not know about well. I want labs to focus more on that.
@donyenrique14@61_rain@MizoChris If you have full chip, any chip with minor defects will have to be thrown out. If you use that same chip and cut 2 compute units off, you can increase your effective yields because not as many dies will have to be thrown out saving money.
@ItsmeAjayKV I think qwen 4 is gonna be overhyped honestly, I think it’s more important to focus on general purpose improvements for qwen 4 rather than focusing on coding ability improvements, I mean engram will likely give a boost by offloading simple words from the attention layers anyways.
@ItsmeAjayKV@QwenDevs Is it actually confirmed to be 35b or no, because a3b is just active parameters. It doesn’t necessarily mean the model will be 35b