Cała rodzina 5.6 to znaczący skok w pracy agentowej. Nawet model Luna potrafi godzinami pracować nad prostymi iteracjami takimi jak pilnowanie kolejki merge train i delegowaniu fixów do subworkerów.
Jeśli u Ciebie nie działa, zacznij od sprawdzenia globalnych i projektowych zasad
@mattshumer_ Strong recommendation: set up deterministic hooks and safeguards against destructive actions, and never let AI execute sudo commands without explicit authorization (e.g. Touch ID)
Strong recommendation: set up deterministic hooks and safeguards against destructive actions, and never let AI execute sudo commands without explicit authorization (e.g. Touch ID)
@hqmank I've been using YOLO mode in both Codex and Claude Code every day for about 4 months, but today I implemented a deterministic harness for destructive actions like this one.
@rezoundous GPT-5.6 sol as orchestrator works perfect. It handle long running tasks without effort. Combine this with Grok 4.5 xhigh to lower your token usage.
@haider1 On narrow, well-defined tasks, Luna genuinely surprises. But juggle 6 things at once, and its focused intelligence just can't match Sol's versatility.
@JaydenDavisNC Even if they pull it now, it'll be back - it's only a matter of time.
Most people aren't ready to use it through the API; it's just not accessible enough yet. And Anthropic needs our data to keep improving their products."
@haider1 Never underestimate Google. Sleeping giants are most dangerous right before they wake up. They don't even need to win the AI race - just put decent AI on every android device, no data center required.
@Angaisb_ Would you prefer that Sol Ultra wasn't available in the Plus plan? IMO it's still a great tradeoff - you can test the full model offering, but you must be aware of your plan's limitations.
@elonmusk Healthy competition never killed anybody. Since I tried Grok 4.5 for the first time, it's become my new number one for code implementation. However, as an agentic model, it's still far behind OpenAI & Anthropic.