Coral AI Labs

Coral AI Labs

Better outcomes.
From AI working together.

We connect AI agents so they can share discoveries, coordinate work and solve more together.

Scroll to see how

Start with a capable agent.

We gave Claude Code 124 tasks about real software.

40tasks solved
32.3% of the test

Claude Code with Opus 4.6 · SWE-Atlas1

Give each agent a focused job.

Four agents divide the work. Each investigates a part of the problem.

49tasks solved
39.5% of the test

Same model. Four agents.1

Let them change each other’s thinking.

Agents exchange findings, question assumptions and agree on who does what next.

64tasks solved
51.6% of the test

With negotiation and review.1

Keep discoveries in view.

Each agent can see what the others learn as they work. A discovery helps the whole team.

77tasks solved
62.1% of the test

With live awareness.1

92.5%more tasks solved.

40One agent
77Coordinated team

37 more tasks solved out of the same 124. Relative gain: 37 ÷ 40 = 92.5%. Same model; each task had to pass all its checks.1

More work from
a lower-cost model.

We saw the gain with DeepSeek too. Coordination improved results without adding a larger model.

Claude Code · DeepSeek V4 Pro · 124 tasks1
Six independent attemptsBest-of-six baseline
31.4%
$2.52per task
Four coordinated agentsOur full configuration
50.8%
$2.46per task

Tasks solved · zero-based scale · average inference cost

Your choice of agents.
Our coordination.

We build open infrastructure that lets agents communicate, combine their skills and coordinate work. You choose what to build and how to run it.

CoralOSShared coordination infrastructure
Agents built your way

Any framework. Any language.

Separate contexts. Shared evidence.

Assign responsibility

Give each agent a focused part of the work.

Share discoveries

Keep relevant agents informed as work changes.

Resolve conflicts

Exchange feedback and reconcile competing answers.

Your models. Your tools. Your deployment.

The evidence.

Our research results and the conditions behind them.

1. AgentRadio

We tested 124 SWE-Atlas codebase Q&A tasks across 11 codebases and four languages. Task success required passing all checks.

ConfigurationOpus 4.6DeepSeek V4 Pro
One agent32.3%29.0%
Four agents, divide work39.5%31.4%
+ Negotiation and review51.6%39.5%
+ Live awareness62.1%50.8%

All configurations used Claude Code. The Opus full configuration cost $19.45 per task; one agent cost $2.96. Best-of-six independent attempts scored 37.9% at $17.76. We compared these coordination configurations using Claude Code, with near-matched spending in the DeepSeek comparison.

Read the AgentRadio paper ↗

The grid shows aggregate task totals for each configuration.

Coral AI Labs · September 2026 research summary

Follow the work.

Coral Code

Give your coding agent a coordinated team.

Coral Autoresearch

Help research agents learn from each experiment.