GPT-6 Astra vs DeepSeek V4.1 on an interactive koi pond
Same koi-pond prompt: Astra finished in 20 minutes for $7, DeepSeek in 1h40m for $0.90; Astra had finer detail, DeepSeek a richer environment but weaker koi.
GPT-6 Astra vs DeepSeek V4.1-1017 Flash on the same interactive koi pond. Same prompt, max reasoning > Astra: 20 minutes, $7 > DeepSeek: 1 hour 40 minutes, $0.90 The math: Astra is 5x faster, DeepSeek is about 8x cheaper. What the developer saw: > Astra nailed the details and came out more original than the usual GPT output > DeepSeek put more work into the environment, and it looks stunning > DeepSeek's koi came out weaker My read on when to pick which: > Deadline or fine details: Astra > Tight budget and you can wait: DeepSeek, then patch the weak parts yourself One run each, so treat it as a data point, not a benchmark.
This post quotes another post: x.com/cashhheth/status/2091158897008476395