Platform to build and run AI agents for go-to-market teams, working from the CRM, email and call data a team already holds. No pricing is published.
SwarmBench by Acyclic Labs
A benchmark, open-source infrastructure and RL environments from Acyclic Labs for measuring and training how well AI models coordinate thousands of agents working on one task.
- Artificial Intelligence - AI
- Machine Learning
TrustList summary
Not yet independently verified. Dated from the Y Combinator launch post of 13 November 2026; the maker's site gives no launch date, pricing or licence, and SwarmBench results are the maker's own published findings. We will update this when it can be confirmed, and remove this note.
Running ten thousand agents on one job sounds like a shortcut to finishing hours of work in minutes. Acyclic Labs, a Y Combinator company, found that it often is not: in its tests, models coordinated so poorly that a swarm could do worse than a single agent. SwarmBench is its answer, a benchmark of the coordination itself.
Measuring delegation and speculation
SwarmBench measures how well a model hands tasks to other agents and how well it speculates, meaning it anticipates upcoming work and starts agents on it early. The evaluations separate these coordination bottlenecks across swarm architectures and model families. The company's first public findings say more agents is not always better: performance depends on the type of workload, and coordination failures can wipe out the gains of scale. The results are published on its site.
Infrastructure from a laptop to a datacenter
Acyclic began with open-source infrastructure that it says can run 10,000 agents in a single session and put a datacenter's worth of compute behind one conversation within seconds. It also builds traces and reinforcement learning environments with long, many-agent tasks, and has begun training models, both frontier models and its own, to work together better. Its loop is stated plainly: measure, train, repeat.
Who it is for
The company addresses model labs, teams building multi-agent systems, partners and researchers. Its founders, Ram Vinjamuri and Varun, studied at Cambridge; Varun previously shipped enterprise multi-agent systems at Palantir and Tracelight.
Key facts
Maker: Acyclic Labs Launched: Y Combinator launch post, 13 November 2026 Products: SwarmBench, agent swarm infrastructure (CLI on GitHub), coordination RL environments Pricing: not published Backed by: Y Combinator
Categories & features
- Artificial Intelligence - AI
- Machine Learning
Request a demo or quote from SwarmBench by Acyclic Labs
Reviews
Write the first review of SwarmBench by Acyclic Labs
Used it? Your experience helps other buyers decide.
Write a reviewQuestions & answers
No questions yet. Be the first to ask about SwarmBench by Acyclic Labs.
SwarmBench by Acyclic Labs alternatives
Similar listings buyers compare against this one.
Workspace where a squad of AI agents takes tasks from a board and returns reports. Early access price $49 per month, listed at $99, per its pricing page.
AI agents that automate back-office work such as purchase order entry and invoice matching across the tools a team already uses. Plans start at $450 per month billed yearly, per its pricing page.
Memory API for AI agents that returns current facts with dated sources, filtered to each user's permitted view
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.