GPT-6 Astra vs DeepSeek V4.1 Flash in Three.js. Same kind of task, two completely different approaches. The visual difference alone is pretty interesting. Source footage and original material: OpenDesign

@OpenDesignHQ how is the quality this close??? you should of put the price difference to, this is INSANE, i was starting to think good ai was gonna be only for the rich but i dont think that at all anymore.

@OpenDesignHQ you forgot to put the prices XD

@OpenDesignHQ What? People were showing Astra to be doing AAA visuals

@OpenDesignHQ prompt or it didn't happen.

@OpenDesignHQ 😭😭😭😭 istg i am testing it too feels like a pro model too damn fast nad just works

@OpenDesignHQ Basically the lesson here is to get DeepSeek to do the initial implementation and then Astra to raise the fidelity.

@OpenDesignHQ What i find most interesting/exciting is a cheap, open source flash model doing better than closed frontier from just a few months ago. Capabilities increasing maybe a bit superlinearly but cost of intelligence in exponential decay

@OpenDesignHQ Gemini

@OpenDesignHQ you can totally get either of them to produce either visuals you only need to be specific about what you want, otherwise the model will just choose their own way of implementing stuff..

@OpenDesignHQ astra visuals hit different ngl

@OpenDesignHQ i hate how monkeys pretend like reasoning levels don’t matter so they don’t mention it at all

@OpenDesignHQ what's the prompt ？

@OpenDesignHQ which reasoning level was astra on? after today's xhigh vs medium quota thing that setting changes the comparison a lot

@OpenDesignHQ The practical test is iteration: at ~$0.03 vs ~$0.59, Flash buys teams roughly 20× more retries. That changes the workflow—not just the benchmark.

@OpenDesignHQ @rafafalleiros

@OpenDesignHQ They kinda look the same. Weird

@OpenDesignHQ Astra es mejor siempre

@OpenDesignHQ Pure cinematic brilliance.

@OpenDesignHQ 오호

@OpenDesignHQ damnn

@OpenDesignHQ I think the best setup for now is astra making orchestrator then deepseek v4.1 as a sub agent..

@OpenDesignHQ side by side clips are fun but one run each is vibes. run the same prompt 10 times and show the misses too, the fails are the interesting bit

@OpenDesignHQ Does DeepSeek has such nice subscriptions or is it API-only?

@OpenDesignHQ Cost difference?

@OpenDesignHQ astra visuals hit different ngl

@OpenDesignHQ GPT-6 Astra is chasing shader complexity. DeepSeek V4.1 Flash is chasing draw-call efficiency. Same Three.js scene, different GPU trade-offs.

@OpenDesignHQ Considering it's 7% in terms of costs, I think it's pretty clear who won this.

@OpenDesignHQ DeepSeek implemented enemies correctly while Astra visuals better

@OpenDesignHQ 3D aside, cannot believe that deepseek robs Astra on UI. Another funny thins is that Astra still does the same (section title) it learned somewhere around gpt 5.3 timeframe.