$0.39 vs $11.42 per task. 47.3 vs 61.7 on Coding Agent Index v2.0.
Codex + DeepSeek V4.1 Flash vs Claude Code + Opus 5.5, both on Artificial Analysis' cost frontier.
~29x the API cost for +14.4 points.
Still my split: Claude for business code, DeepSeek as fallback.
Where's your cutoff?
@kimmonismus The receipt thing matters more than people think. I've been burned by AI summaries that quietly dropped the actual commitment. Linking back to source is the only way to verify before acting on a reminder.
@TheAhmadOsman I grabbed a 4090 before prices went stupid and it's paid for itself twice over. The real trick is picking workloads that actually fit the VRAM you've got instead of chasing the biggest model.
@poteto@shaoruu X group chats don't have bot APIs yet, but I've been using a webhook bridge to relay messages into Slack and it works fine for now. Just watch out for rate limits if you're sending a lot.
@theo Good catch. I hit the same thing last week and thought I was going crazy watching the 5h meter drop way faster through the SDK. Switching back to Claude Code directly for now until they patch it.
Anthropic's Applied AI team: one agent with skills often beat a stack of subagents. One customer's single loop ran ~2x faster.
The split I still plan: a fresh reviewer for my PR agent. Their guide keeps verification as a valid one.
Which subagent would you fold first?