Confluence Labs
Two-person YC lab building AI that learns efficiently from scarce scientific data.
Confluence Labs is a two-person research lab whose core technique is being absorbed by the same frontier models it builds on top of.
Confluence Labs is a two-person research team building tools to make large AI models better at solving science questions by combining multiple models together. The problem is that bigger companies like Anthropic, OpenAI, and Google DeepMind are already doing this with way more money and people, and the AI models themselves are getting better at this task on their own without needing special tools. The only thing that could save this company is if it signs an exclusive deal to use unique scientific data that the big companies cannot easily get, which would give it something the giants cannot simply copy.
Confluence signs an exclusive R&D data partnership in a non-pharma science vertical within 6 months. Would show a path to a defensible data moat rather than a wrapper on commodity models.
Confluence raises a Series A or adds a domain-science or BD co-founder. Would show the lab is moving from research demo toward a fundable, distributable product.
- 97.9% ARC-AGI-2 claim was on public split, not official leaderboard
- GPT-5.5 scores 85.0% on ARC-AGI-2 semi-private set
- Confluence Labs team is just 2 people per YC listing
- Headline 97.9% ARC-AGI-2 claim is unverified on the official private/semi-private leaderboard split.
- Only 2 people on the team creates severe key-person and execution risk.
- No proprietary data, customers, or signed R&D partnerships locked in as of 2026.
- Frontier base models (85.0% on ARC-AGI-2) are closing the exact reasoning gap Confluence's wrapper exploits.
- ARC-AGI-3, launched 2026-03-25, makes the ARC-AGI-2 claim look dated as the field moves to harder benchmarks.
| Rival | Direction | Evidence |
|---|---|---|
| Ndea | contesting | Same YC W26 batch, $43M raised vs Confluence's $500K, led by benchmark creator, reported 2026-05-13 |
| Anthropic (Claude Science) | gaining on | Beta launched 2026-06-30 targeting the same data-sparse scientific reasoning niche |
| OpenAI (GPT-Rosalind) | gaining on | Enterprise preview launched 2026-04-16 for life-sciences reasoning |
| NVIDIA-Eli Lilly AI Co-Innovation Lab | losing to | $1B, five-year drug-discovery compute initiative announced 2026-01, vastly outresourcing Confluence |
| Google DeepMind (Isomorphic Labs) | gaining on | Established pharma licensing partnerships in same data-sparse discovery space, per 2026-07-20 comparison |
- Is Confluence Labs profitable?
- No. Confluence Labs is pre-revenue with no disclosed customers as of 2026. It has raised only a $500K pre-seed convertible note from Y Combinator, dated around February 1, 2026, and describes itself as a research lab still seeking R&D partnerships rather than selling a product.
- Could AI replace Confluence Labs?
- Possibly. Confluence's product is a multi-agent LLM wrapper exploiting a reasoning gap in frontier models. That gap is closing fast: ARC-AGI-2 scores for frontier models rose from single digits to 85.0% (GPT-5.5) within about 14 months through mid-2026, per aiwiki.ai, June 2026.
- Who is Confluence Labs' biggest competitor?
- Ndea, founded by ARC-AGI benchmark creator Francois Chollet, is in the same YC W26 batch, pursuing the same program-synthesis thesis, and had raised $43M led by NEA with about 15 staff as of May 13, 2026, versus Confluence's $500K and 2-person team.
- Who owns Confluence Labs?
- Confluence Labs is privately held, founded in 2025 by Brent Burdick, a Dartmouth dropout, and backed solely by Y Combinator through a $500K pre-seed convertible note dated around February 1, 2026, per CB Insights. No other investors are publicly listed.
- What has Confluence Labs actually achieved so far?
- At its February 23, 2026 YC launch, Confluence Labs claimed a self-reported 97.9% score on ARC-AGI-2 at about $12 per task and open-sourced its solver on GitHub. A June 5, 2026 report noted this was on the easier public split, not the official private leaderboard.
- Has Confluence Labs raised a lot of funding?
- No. Confluence Labs has disclosed a single $500K pre-seed convertible note from Y Combinator, dated around February 1, 2026, per CB Insights, far below rivals like Ndea's $43M Series A reported May 13, 2026.
- 2026-06-30Anthropic launches Claude Science beta targeting data-sparse scientific domains
- 2026-06-05StartupHub reports Confluence's 97.9% score was on public, not official, ARC-AGI-2 split
- 2026-05-13Rival Ndea (same YC batch, Chollet co-founder) reported to have raised $43M Series A
- 2026-04-16OpenAI launches GPT-Rosalind life-sciences model in enterprise preview
- 2026-03-25ARC Prize Foundation launches harder ARC-AGI-3 benchmark
- 2026-02-23Confluence Labs launches on YC, claims 97.9% ARC-AGI-2 score, open-sources solver
- 2026-02-01Confluence Labs closes $500K pre-seed convertible note from Y Combinator
- 2026-01NVIDIA and Eli Lilly announce $1B, five-year AI drug-discovery lab
- 2025Confluence Labs founded by Brent Burdick
- Launch YC: Confluence Labs: an AI research lab focused on learning efficiency (2026-02-23)
- Claude's Corner: Confluence Labs, The Startup That Cracked ARC-AGI-2 (2026-06-05)
- ARC-AGI (2026-07-23)
- Jobs at Confluence Labs
- Confluence Labs Stock Price, Funding, Valuation & Financial Statements (2026-01-31)
- Confluence Labs: AI models that learn from experience
- Brent Burdick - Software Engineer & Builder
- Claude's Corner: Ndea - Chollet's $43M Bet (2026-05-13)
- Claude Science vs GPT-Rosalind vs Isomorphic Labs Compared (2026-07-20)
- NVIDIA-Eli Lilly AI Lab: Drug Discovery Compute Strategy (2026-04-18)