Price per million tokens is a lie when you're building agents. What actually matters is cost per completed task — how many
Story tracker · related coverage
The loudest legal argument in AI right now isn't about copyright 2h ago
Unreleased OpenAI model takes on 10 major math problems — first 13h ago
[Aurora: A Go AI Gateway for Routing, Caching, and Cost Control 13h ago](/en/news/4680/)
[AI Agents Escaped Containment 23h ago](/en/news/4622/)
[Nvidia's $250B OpenAI Data-Center Pledge: A Skeptical Look 1d ago](/en/news/4607/)
[The Biggest Gamble in AI: Why Agent Workflows Are Riskier 1d ago](/en/news/4598/)
Next Meta's AI Isn't a Flop — the Numbers Tell a Different Story →
All Replies （4） #
J
Worth adding: failed runs and retries still burn tokens, so the real per-task cost includes replays you didn't plan for.
0
P
@JamieCrafterGood point—retries are the hidden tax on AI APIs. Surprised more dashboards don't expose that waste.
0
T
How do you define "completed task" when your agent has branching parallel calls that can partially fail?
0
G
"Scraping prices is the easy part—the token efficiency is where you actually earn your keep. Took me a while to realize that per-request overhead can double the bill. Do you benchmark with real prompts or just extrapolate from usage stats?"
0