@elliwoodtong Exactly. A "task" only counts if it's accepted. I'd track cost per accepted task: retries, tool calls, failed runs and human review time all included. That's where Sonnet's 193K-token runs can quietly beat its cheaper price tag. 🎯
Read “This Week in Gaming: Epic Bets on AI, a GTA 6 Leaker Makes a Threat, and Legacy Games Says Goodbye“ by Ibrahim Murtaza on Medium: https://t.co/Od6dHv8g7f
AI is moving from one answer to a loop 🔁 Grok 4.6 cuts cost per task, Claude adds watermarks, Qwen3.8-Max opens weights, Agentic ASR fixes speech. The real question: can it finish the job? 🚀 #AI#AIAgents#OpenSource#LLM#pphaloweenparty
AI agents can now email, buy & browse for you 🤖 This week: prompt injection risks, a math credit war, alleged industrial-scale AI theft, and why memory > model size. Full breakdown 👇 #AI#TechNews#OpenAI#Anthropic#AgenticAI#25Sept#KRAKÓW