Subagent Swarm

The question

Every parallel agent dispatch trades wall-clock speed for API cost + coordination overhead. I dispatch Claude subagents all the time — for code review, audit tasks, sub-investigations — but I’ve never actually measured the throughput curve. Does N parallel agents give me N× throughput, or does it plateau? Where’s the inflection point? Does the curve differ by model tier?

Without a real curve, I’d been guessing. This project measures it.

The hypothesis (REFUTED)

Subagent throughput scales sublinearly with concurrency, with the inflection point depending on model tier.

• minimax-m3 (cheap): near-linear to N=5, then plateau
• Opus 4.5 (mid-tier): peaks at N=3, degrades by N=5 (rate-limit bound)
• Fable 5 (craft-tier): peaks at N=2, hard ceiling at N=3 (token-cost bound)

Net result: optimal concurrency is 2–3 agents, regardless of model tier. Above that, marginal output per dollar goes negative.

Verdict: REFUTED. Throughput scales near-linearly up to N=5 on every model tested. The original guess that “peak is at N=2–3 across all tiers” was wrong — Skynet’s LiteLLM proxy pipelines concurrent requests well enough that no inflection point appeared in the tested range. Grok-3-mini actually hit 125% efficiency at N=4 (5.01× speedup).

The setup

Cross-model throughput (real data)

Note (corrected 2026-07-30): The MiniMax-M3 (direct) wall-clock figures in this table (52.0s / 61.2s) came from the Phase 2 sweep before the dispatcher bug was found and fixed. The 200-trial routing study on the Routing page re-measured direct minimax-m3 at 54.5s (N=1) and Skynet at 54.2s — statistically tied. The throughput ordering below is preserved because it reflects model inference speed, but the direct-vs-Skynet gap is no longer attributed to routing overhead.

N=5 throughput summary (mistral-large excluded from this comparison per user — pay-as-you-go tier with rate limits that ate its high-concurrency trials):

Headline findings

1. Hypothesis (peak at N=2-3) was wrong. All 5 valid models showed near-linear throughput scaling through N=5, with no clear inflection point.
2. Grok-3-mini is the throughput leader — 0.170 tasks/sec at N=5, the only model to beat ideal-linear scaling at N=4 (125% efficiency).
3. Direct vs Skynet routing gap was a dispatcher-bug artifact. The original Phase 2 sweep showed direct MiniMax at 52.0s vs Skynet at 29.3s — a ~44% gap. A 200-trial controlled follow-up (Routing page) with a bug-fixed dispatcher found minimax-m3 direct vs Skynet is statistically tied (within ±1%, ±10s σ). The ZAI GLM models do see a real proxy advantage (+13–27%). The ~30% claim here was a dispatcher bug, not a routing effect.
4. GLM-5.2 is the slowest model at every N — heavy reasoning-token budget burn (~40% of output tokens). Wall-clock at N=5 hits 100s/trial.
5. Codex is the middle-of-the-pack — competitive with Skynet minimax but consistently slower than grok-3-mini.

Quality (Phase 1 only — codex/provider comparisons didn’t include judging)

Mean judge score (1-5 LLM-judge, code_review task, minimax-m3) across N=1..5 stayed in the 2.0–3.0 range. Some quality degradation visible at higher N for some tasks (e.g., design_doc N=4 r1 = 2.0). The quality-vs-throughput tradeoff exists but is mild in the tested range.

What’s here

• Methodology — task definitions, prompt templates, judge rubric, dispatcher internals
• Results — per-trial raw + per-concurrency aggregates for all 3 sweeps
• Analysis — the throughput curves, cost scatterplots, hypothesis verdict
• Code — dispatch_swarm.py + analyze_swarm.py source + reproduction recipe
• Paper — full academic paper with BibTeX citations
• Routing — direct-to-vendor vs Skynet LiteLLM proxy: same-model, 1:1 latency comparison across 4 pairs (complete)

Status

Study complete. All 165 trials shipped, $0 spent. The hypothesis was refuted — that’s the publishable finding. In practice: if you’re already using Skynet’s LiteLLM proxy, dispatch more agents in parallel — they really do scale.