---
type: "raw_signal"
id: "scaling-agent-teams-reach-the-frontier-at-30-648e89"
slug: "scaling-agent-teams-reach-the-frontier-at-30-648e89"
title: "Scaling agent teams: reach the frontier at 30% of the cost"
description: "Cohere published a controlled study of coding-agent team design using ProgramBench, a benchmark of 200 programs rebuilt from behavior."
retrieval_nugget: "Cohere published a controlled study of coding-agent team design using ProgramBench, a benchmark of 200 programs rebuilt from behavior."
observed_at: "2026-10-14"
record_date: "2026-10-14"
date_kind: "observed_at"
why_it_matters: "Cohere published a controlled study of coding-agent team design using ProgramBench, a benchmark of 200 programs rebuilt from behavior. The company reports that well-organized teams improved performance on challenging tasks by up to about 10 percentage points over single agents, at 3–7 times the cost. In one comparison, a Sol lead coordinating 16 lower-cost Luna planning workers achieved 61.8% rewa"
novelty: "notable"
verification_level: "source-linked"
signal_type: "research"
source_platform: "cohere.com"
topics: []
entities: ["Cohere","ProgramBench","Sol","Luna"]
related_patterns: ["agent-economics-moves-to-completed-work","harness-architecture-outlives-model-choice","verification-bandwidth-is-the-scarce-resource"]
source_url: "https://cohere.com/blog/lessons-from-scaling-teams-of-frontier-coding-agents"
source_urls: ["https://cohere.com/blog/lessons-from-scaling-teams-of-frontier-coding-agents"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "signal:scaling-agent-teams-reach-the-frontier-at-30-648e89"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/signals/scaling-agent-teams-reach-the-frontier-at-30-648e89/","markdown":"https://newruntime.com/signals/scaling-agent-teams-reach-the-frontier-at-30-648e89.md","json":"https://newruntime.com/signals/scaling-agent-teams-reach-the-frontier-at-30-648e89.json"}
---

# Scaling agent teams: reach the frontier at 30% of the cost

## Retrieval answer

Cohere published a controlled study of coding-agent team design using ProgramBench, a benchmark of 200 programs rebuilt from behavior.

Cohere published a controlled study of coding-agent team design using ProgramBench, a benchmark of 200 programs rebuilt from behavior. The company reports that well-organized teams improved performance on challenging tasks by up to about 10 percentage points over single agents, at 3–7 times the cost. In one comparison, a Sol lead coordinating 16 lower-cost Luna planning workers achieved 61.8% reward, versus 63.2% for an all-Sol team delegating planning, implementation, and verification. The reported cost fell from $510 to $155. Cohere also says delegating exploration and planning alone matched the performance of delegating all three stages at about half the cost. These are results from the study's harness and benchmark, not evidence that the same allocation will be optimal for every coding workload. The central finding is that where teams spend additional inference compute can matter more than simply adding agents.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Raw Signals](https://newruntime.com/signals.json): Open every source-linked raw signal.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)