“`html
A recent post on Reddit suggests that coding is now effectively automated for the majority of tasks, particularly those in the “boring 90%” category. This claim comes from a user who managed to refactor a substantial FastAPI service with minimal human intervention and at a very low cost.
- The refactoring involved mass edits across over 120 files, requiring approximately 400 steps and consuming around 2 million tokens, all for just $3. The AI system handled the task without any manual input from humans.
- Two models—DeepSeek v4 and Hunyuan Hy3—were used as cheap workers in this process. These models were approximately 80 times cheaper than a model called Opus, which is often considered a standard reference point for such tasks. The cost breakdown showed that the active parameters of both models (21 billion) are significantly lower than those of Opus.
- The performance was noteworthy as the open-weight tier responded faster than Opus, leading to most tasks completing in under an hour. However, there were some delays with more complex steps or escalations that took longer.
While this news highlights significant progress in automating routine coding tasks, it also underscores the ongoing need for models like Opus, which appear capable of handling even more challenging and varied workloads. The comparison between different AI models further emphasizes the importance of choosing the right tool for specific applications.
“`
“`md
* Coding is now effectively automated for a majority of tasks, particularly those in the “boring 90%” category.
* A user managed to refactor a substantial FastAPI service with minimal human intervention and at a very low cost ($3) using two models (DeepSeek v4 and Hunyuan Hy3), which were approximately 80 times cheaper than Opus. The active parameters of both models are significantly lower, highlighting their efficiency.
* There was notable performance disparity; the open-weight tier responded faster than Opus but still took longer for more complex steps or escalations.
Takeaways:
– Coding automation is advancing rapidly, especially for routine tasks.
– Models like DeepSeek v4 and Hunyuan Hy3 show promise in cost-effective AI solutions for coding refactoring.
– The choice of model matters; the open-weight tier performed better but still required some manual intervention compared to Opus.
“`
Originally published at reddit.com. Curated by AI Maestro.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
