REDDIT · REDDIT// 28d agoNEWS

Grok 4.20 multi-agent beta burns 333k tokens on trivial prompt

A user testing Grok 4.20 Multi-Agent Beta via OpenRouter triggered 333,210 input tokens and 53,859 output tokens — costing $0.47 — just by asking how many agents were working on a response. The token explosion on a near-zero-complexity prompt raises questions about whether xAI's multi-agent coordination overhead is a bug or an inherent architectural cost.

// ANALYSIS

This is a red flag for anyone considering Grok's multi-agent tier in production: the cost-per-query economics appear broken at launch.

–333k input tokens on a three-word follow-up question suggests the system is passing full conversation context to every spawned agent simultaneously — a naive fan-out architecture
–At that rate, real workloads could cost 10-100x more than equivalent single-agent calls with no clear quality benefit
–The original poster noted Grok 4.2 single-agent scored reasonably on the AA-Index, making the multi-agent overhead even harder to justify
–OpenRouter users are effectively beta-testing internal xAI orchestration plumbing — pricing and behavior may change without warning
–This mirrors early complaints about other multi-agent frameworks (AutoGen, CrewAI) where agent-to-agent context passing silently multiplies token costs

// TAGS

grokllmagentinferencebenchmark

DISCOVERED

28d ago

2026-03-15

PUBLISHED

28d ago

2026-03-14

RELEVANCE

6/ 10

AUTHOR

rnahumaf