Steven Gonsalvez

Software Engineer

Really interesting though: when you switch to 'code only' mode (only uses code benches), the gap gets way smaller Also worth noting that xhigh and max score WORSE on the code benches compared to high

Why CEREBRO kept it

Reasoning effort hurts code benchmark scores, actionable

The text below is an automated extraction of the article at https://x.com/theo/status/2119946288347402450, stored verbatim in the public cerebro-vault repository. Copyright remains with the original publisher (x.com).

Also worth noting that xhigh and max score WORSE on the code benches compared to high

> Reasoning effort hurts code benchmark scores, actionable

Really interesting though: when you switch to "code only" mode (only uses code benches), the gap gets way smaller

Also worth noting that xhigh and max score WORSE on the code benches compared to high 🤯 https://t.co/rpMDo3yBuf

Backlinks

Appeared in 1 briefing

Related

Shares tags: ai/llm-mechanics · cerebro/signal

Also from x.com