I must be doing something wrong. I gave v4 pro a try a couple of days ago, gave it a simple prompt like "clean up functions x and y in file z" and it would always start off promising, just to quickly get sidetracked, start hallucinating problems in the code, and just get stuck for hours until I interrupt it:

— hmm — 0x2D696370 — little-endian bytes: 70 63 69 2D = 'p','c','i','-' — hmm — WAIT — WAIT — !!!!! — *WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — *HOLD ON — HOLD ON — HOLD ON — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — !!!!!!!! — *WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — *OK — WAIT — I THINK I FINALLY SEE THE WHOLE PICTURE — I NEVER READ IT — AND — THE LAYOUT — hmm — !!!!! — *WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — WAIT — HOLD ON — HOLD ON — HOLD ON — HOLD ON

Then gave the same to Sonnet 5 and it was done 15 - 30 minutes later. I tried v4 pro both in claude code and codewhale with similar results. Haven't tried the new deepseek harness.

I've been using the 0731 Flash V4 model, via Opencode, and I've had no major issues. It feels very comparable to Opus 4.6/4.7, that I use at work. I haven't ran into any of the problem you mention, so that might be a quirk of V4 pro, the specific harness, or maybe the host you're using?

I guess this was a heavily quantized version from openrouter? I've never had that experience in the last months of quite intensive use of the official deepseek api.

I used flash with pi and it worked pretty well.

It built this whole IaC plugin from scratch: https://github.com/fllstck/nebius-alchemy