deepseek-v4.1POST
V4.1 Flash 1017 feedback: shorter outputs and better agent scores, but cache misses still cost on long runs
The author credits shorter outputs and higher agent scores, but notes long agent runs reuse prefixes and still pay prefill on cache misses, recommending exact caching and hard tool budgets.
Milon
@milonspaceArchived 11 days agoOriginally posted 2026.10.18
Milon @milonspaceOriginally posted 2026.10.18
DeepSeek-V4.1-Flash-1017 is getting all the attention. Shorter outputs, same price, better agent scores. That's real progress. But it doesn't fix the loop. Long agent runs reuse most of the same prefix. When the cache misses, you pay for prefill again, regardless of how short the next answer is. Upgrade the harness alongside the model. Exact caching. Hard tool budgets. Stop when the budget runs out. The SVG will look fine either way. The session bill won't.
x.com/milonspace/status/2111508553005666421Archived 2026.10.18 16:39