All posts
Blog

How Top VC Firms Evaluate Founders: The Taste Matrix of Sequoia, Benchmark, a16z, Founders Fund, Thrive, and Lightspeed

We analyzed roughly 70 seed and Series A checks written between June 2024 and June 2026 by six of the most influential venture firms in the world. Each firm treats one signal as a hard gate, and the gates barely overlap. Six founder personas tell you in minutes which firm is structurally built to fund you.

Every top VC firm says it backs great founders. The data says each one means something different

Ask six tier-1 venture firms what they look for in founders and you will hear six versions of the same phrase: exceptional people, huge markets, early conviction. The words are interchangeable. The checks are not.

At Vela, a leading tier-1 quant VC, we treat investment taste as data. So we analyzed roughly 70 seed and Series A rounds announced between June 2024 and June 2026 across six of the most influential firms in venture: Sequoia Capital, Benchmark, a16z, Founders Fund, Thrive Capital, and Lightspeed. For each check we reconstructed the founder's background, the team composition, the sector, the traction at the time of investment, and how the founders met their investors. Then we scored each firm on eight signal dimensions.

6firmsSequoia, Benchmark, a16z, Founders Fund, Thrive, Lightspeed
~70roundsSeed and Series A checks analyzed, June 2024 to June 2026
8signalsDimensions scored 0-10 per firm, from exits to GP relationships

The headline finding: each firm treats exactly one signal as a hard gate, a signal whose absence disqualifies, and the gates barely overlap. A founder who trivially passes Founders Fund's gate may not clear Benchmark's, and vice versa. “Fundable” is not a property of the founder. It is a property of the founder-firm pair.

No two firms are pricing the same founder. Each has one signal it treats as a gate, and across six firms the gates barely overlap.

Which founder are you? Six personas, matched to the firms that fund them

Before any firm-by-firm detail, start here. Every founder in the ~70 analyzed rounds fits one of six personas. Find yours, and you immediately know which firms are structurally built to fund you and which will pass no matter how good the deck is. Green chips are your primary targets.

P1The Frontier Researcher

You invented something the field now builds on. You may have no product, no revenue, and no company yet.

  • PhD or tenure at MIT, Stanford, Berkeley, or CMU, or a top-lab badge from OpenAI, DeepMind, Anthropic, or Meta FAIR
  • Papers or systems other researchers cite and extend, like AlphaChip, ESM1, or a decade of FAIR robotics work
  • Your co-founder is a research equal or a practitioner who lived the problem firsthand
Sequoiaa16zLightspeedThrive

Lead with the research artifact, not the market size. Sequoia wrote first checks into Ricursive and Traversal with zero product. Revenue slides are wasted pages for you.

P2The Traction Outlier

No pedigree, no famous lab, maybe no degree. But the market already said yes.

  • $1M+ ARR, ideally self-funded, or profitability before your first institutional round
  • A growth curve that happened without paid marketing or a brand-name investor
  • Age and resume are irrelevant: Mercor's founders were 21 with $1M ARR
BenchmarkLightspeed

Lead with the revenue chart. Benchmark's entire 2024-2026 seed set had adoption proof before the check. HeyGen was profitable and near $35M ARR when Benchmark engaged.

P3The Open-Source Maintainer

A community already depends on your code. The company is just the commercial wrapper.

  • A repository with tens of thousands of stars, or infrastructure running in production at scale, like vLLM on 400K GPUs
  • Thousands of contributors, or adoption inside enterprises and hospitals before incorporation
  • Your founding team is the maintainer group itself
a16zBenchmark

Lead with adoption metrics: stars, contributors, production deployments. For a16z the community is the proof of concept, and for Benchmark it is a revenue proxy (LangChain had 84K stars at check).

P4The Prodigy

You were world-class at something measurable before you could legally rent a car.

  • IOI, IMO, or ACM-ICPC medals, ideally gold: the Cognition team held 10 IOI golds collectively
  • A founding team of competition-circuit peers you have known for 10+ years
  • First company, zero revenue, and comfortable saying so
Founders FundBenchmark

Lead with the medal count and the team roster. Founders Fund is the only firm where competition credentials are a stand-alone gate: $21M for Cognition at six months old, no revenue. Benchmark explicitly cites competition wins too.

P5The Big-Tech Builder

You shipped something famous inside a giant, and you left to fix a problem you lived through.

  • Built at scale inside Meta, Google, Stripe, or Palantir: think ESM1, Stripe Link, or Gmail-scale product work
  • A co-founder who has been your friend or classmate for 5 to 15 years
  • Research depth plus shipping record, like NeurIPS publications alongside production ML
ThriveBenchmarkSequoia

Lead with what you built and who you built it with. Thrive backed Chai at a $150M seed valuation with zero revenue because the founders' building record and decade-long bond were the proof.

P6The Repeat Domain Operator

You already sold a company in this exact domain, and now AI makes the 10x version possible.

  • A prior acquisition in the same field you are rebuilding: Granola's founder had two Google acquisitions in learning and productivity
  • Deep operator lineage: the executives, customers, and hiring pool of your domain know you
  • A technical co-founder whose prior artifact was adopted at scale
LightspeedSequoia

Lead with the exit and the domain thesis. Lightspeed's pattern is near-universal: prior acquisition in the exact domain, plus a GP relationship that often predates the company by a decade.

If you match more than one personaStack them. A Frontier Researcher who also has open-source adoption clears both the Sequoia and a16z gates (that is the Inferact profile). A Big-Tech Builder with revenue clears Thrive and Benchmark. Multiple gates mean a competed round, and a competed round means leverage on terms.

The six firms in depth: gate, criteria, and how to get in

Each profile below gives the firm's revealed strategy, its hard criteria from the 2024-2026 data, and the single sentence that tells you whether you fit.

Sequoia CapitalSeed Fund VI · $200M · 8-12 checks/yr

Sources through the Arc program and elite academic networks years before a company exists, then waits for the right researcher to commercialize the thesis. Backed the AlphaChip inventors (Ricursive), Gemini lead researcher Misha Laskin (Reflection AI), and two tenured CMU professors (Skild AI), several with zero product at the time of the check.

Hard gateElite research credential verified by institution: PhD from MIT, Stanford, Berkeley, or CMU is near-mandatory on the research track
Prior exitNot required. Research credentials fully substitute
First-timersYes, the default for the research track
RevenueNot required. Ricursive, Reflection, and Traversal were all $0 at seed
Team shapeResearcher paired with a practitioner who lived the problem. Solo founders rare
Way inThe GP tracks you for years through academic networks before the company exists. Arc is the structured path
You fit here ifYou invented something the field builds on, and you can name the person who lived the problem you are solving.
BenchmarkFund X · ~$425M · 7-15 checks/yr

The only tier-1 fund that waits for the market to speak: nearly every 2024-2026 check followed $1M+ ARR or mass developer adoption. LangChain had 84K GitHub stars, HeyGen was profitable at roughly $35M ARR, and Mercor's 21-year-old founders had $1M ARR fully self-funded before Benchmark engaged.

Hard gateMarket adoption: $1M+ ARR or equivalent developer traction is near-universal in the observed data
Prior exitConditional. Revenue proof waives it entirely
First-timersYes. Mercor's founders were 21, first company
Revenue$1M+ ARR strongly preferred, profitability even better
Team shapeBig Tech engineering core plus a product or commercial co-founder
Way inEvents, retreats, and YC batches. The only firm that explicitly cites competition wins in its own writing
You fit here ifThe market already said yes, in revenue or adoption you can chart, even if nobody knows your name.
a16z$400M seed fund · $90B AUM · 20-40 checks/yr

Backs open-source communities that exist before the company does, then funds the entity that captures them: the vLLM maintainers (Inferact), the Stable Diffusion co-creators (Black Forest Labs), and SSI's $1B seed for former OpenAI Chief Scientist Ilya Sutskever. Check size is unconstrained because the a16z brand itself accelerates adoption.

Hard gateTop AI lab origin for model bets, open-source adoption for infrastructure: the community is the proof of concept
Prior exitConditional above $100M checks. Lab origin replaces it
First-timersYes, at every check size including $1B
RevenueNot required. SSI, Mirendil, and BFL were all $0 at seed
Team shapeFlexible, 2 to 20+ people. Often the maintainer group or lab team itself
Way inOpen-source and top-lab networks, plus START, Speedrun, and CSX as structured entry points
You fit here ifA community already runs your code, or your badge reads OpenAI, DeepMind, or Meta FAIR.
Founders Fund$17B AUM · ~5-10 seed checks/yr

Incubates companies more than it discovers them. The IOI competitive programming circuit is the credentialing system for first-time founders: Cognition's 10-person team held 10 IOI gold medals and got $21M at six months old with no revenue. A parallel channel runs through the Palantir and Anduril operator network (Valthos, Netic, Armada).

Hard gateIOI-tier competition medals or Palantir-tier executive lineage. The only firm where medals stand alone
Prior exitNot required. Cognition's founders had zero exits
First-timersYes for prodigies. Solo founders not accepted
RevenueNot required. Armada raised a $55M seed at $0 revenue
Team shapeCompetition-circuit peers of 10+ years, or an executive paired with an operator
Way inOften incubated inside the Founders Fund network itself: private conferences, Thiel Fellowship cohorts, the IOI circuit
You fit here ifYour teenage self won international medals, or you ran a Palantir-tier organization and your co-founder ran operations beside you.
Thrive CapitalFund X $10B · ~12 checks/yr across all stages

Writes roughly 12 checks per year total and uses seed bets as relationship-incubation vehicles for later $1B+ growth checks: the Chai seed led to Thrive's Series B at $1.3B, Cursor's Series B led to a $900M Series C co-lead. Thrive was the only institutional term sheet OpenAI received in 2022, and it shows up during market stress when others do not.

Hard gateBuilt at scale inside top tech: ESM1 at Meta, Stripe Link as a teenager, NeurIPS plus Google ML
Prior exitNot required. The building track record replaces it
First-timersYes, when domain credentials are deep
RevenueNot required. Chai's seed was $0 revenue at a $150M valuation
Team shapeDomain-expert CEO plus a product-at-scale co-founder who has been a friend or classmate for 5 to 15 years. No cold pairings observed
Way inThe deepest relationship model in venture. Thrive knows its founders years before the round
You fit here ifYou built something famous inside a giant, and your co-founder has been beside you for a decade.
Lightspeed$40B AUM · ~60-100 investments/yr

Backs founders with a prior acquisition or research impact in the exact domain they are applying AI to: Granola's founder had two Google acquisitions before the seed, Skild AI's founders had a decade of FAIR robotics research. The vertical AI thesis is organized by profession (productivity, robotics, healthcare, music, legal), and 11 global offices surface European founders early.

Hard gatePrior acquisition or defining research in the same domain. The pattern is near-universal
Prior exitPreferred (7/10), the strongest exit weighting in the set
First-timersConditional: needs a prior technical artifact adopted at scale
RevenueConditional. Prior acquisitions substitute (Granola was $0)
Team shapeRepeat founder plus a technical artifact creator, or two research equals with a decade of shared work
Way inDecade-long GP relationships. The Granola partner knew the founder ~10 years before the check
You fit here ifYou already sold a company in this exact domain, or your research defined it, and AI now makes the 10x version possible.

The signal-weight grid: eight dimensions, scored 0-10 per firm

Scores estimate the weight of each signal in the investment decision, derived from pattern frequency across observed checks, not from any firm's stated criteria. A 10 means the signal functions as a stand-alone gate; a 1 or 2 means it is not a factor. The green chip marks the firm that weights each signal most heavily. (SEQ = Sequoia, BMK = Benchmark, FF = Founders Fund, THR = Thrive, LSP = Lightspeed.)

Prior exit to a tier-1 acquirer
SEQ4BMK6a16z5FF2THR3LSP7
PhD / elite research credential
SEQ9BMK2a16z8FF4THR7LSP8
Top AI lab originOpenAI, Anthropic, DeepMind, Meta FAIR
SEQ6BMK2a16z10FF4THR8LSP5
$1M+ ARR at time of check
SEQ3BMK9a16z3FF2THR2LSP4
Competition olympiad credentialIOI, IMO, ACM-ICPC
SEQ4BMK7a16z4FF10THR2LSP2
Open-source / developer adoption
SEQ5BMK8a16z9FF1THR4LSP5
Domain operator lineage
SEQ3BMK6a16z4FF8THR7LSP6
GP relationship depth at investment
SEQ8BMK6a16z6FF9THR10LSP9

Two rows deserve special attention. First, the competition olympiad row: Founders Fund is the only firm where an IOI gold medal functions as a stand-alone qualifier, and Benchmark is the only firm that has explicitly cited competition wins in its own investment writing. Second, the GP relationship row: it is the single most consistently high-scoring signal across all six firms. Thrive scores a perfect 10 (it was the only institutional term sheet OpenAI received in 2022), and Lightspeed's Granola seed came after a roughly ten-year relationship between partner and founder. Cold inbound is structurally disadvantaged everywhere.

Where the money goes: sector appetite by firm

Sector preference splits the six firms almost as cleanly as founder taste does. Green chips mark firms with multiple active bets in the sector; gray chips mark occasional or growth-stage-only exposure. If a firm does not appear on a row, it was absent from that sector in the observed data, and pitching into that absence is pitching into a structural no.

B2B AI applicationsThe only sector where every firm has multiple active bets
SequoiaBenchmarka16zFounders FundThriveLightspeed
Frontier AI labs / modelsSSI, Reflection, Thinking Machines, Mistral, OpenAI
a16zThriveLightspeedSequoiaFounders Fund
AI infrastructure / inferenceLangChain, Inferact, Nexthop, Traversal
Benchmarka16zLightspeedSequoiaThrive
Bio / health / scientific AIChai, Isomorphic, Valthos, Phylo, Abridge
a16zFounders FundThriveLightspeedSequoiaBenchmark
Consumer / creator AIHeyGen, ElevenLabs, Suno, Pika, BFL
Benchmarka16zThriveLightspeedSequoia
Defense / physical worldArmada, Anduril lineage, Hadrian, Castelion
Founders Funda16zSequoiaThriveLightspeed
AI hardware / chipsRicursive, Unconventional AI, Substrate
Sequoiaa16zFounders Fund
AI roboticsSkild AI seed and Series A
SequoiaLightspeeda16zFounders Fund

What this means if you are raising

  • Map your persona to the right gate before you pitch. If you have $1M+ ARR and no pedigree, Benchmark is structurally your best-fit tier-1 conversation. If you have a top-lab badge and no revenue, a16z and Sequoia are. If your credential is a prior acquisition in your domain, Lightspeed's entire model is built around you.
  • Start the relationship years before the round. The GP relationship signal scores 6 to 10 at every firm in the matrix. Arc, START, Speedrun, and firm retreats exist precisely because the firms know their own funnels are relationship-gated.
  • Choose your co-founder like it is a signal, because it is. Solo founders clear almost no gate in this data. Thrive's pairs average a decade of prior relationship, and Founders Fund's teams are competition-circuit peers of 10+ years.
  • Do not pitch a firm whose sector row excludes it. Defense founders pitching Benchmark, or consumer founders pitching Founders Fund, are pitching into a structural absence, not a persuadable gap.

The seventh door: how a tier-1 quant VC evaluates you

There is one more thing the matrix reveals: every gate above is a proxy. A medal, a lab badge, an exit, a star count. Each firm picked the proxy its partnership can evaluate by hand, then built a funnel around it. That is why the gates barely overlap, and why strong founders who sit between personas fall through them.

This is exactly the problem quantitative venture capital exists to solve, and it is why Vela operates as a quant VC: a tier-1 firm where the evaluation is done by models, not by a single partner's pattern library. Our AI agents score founders across all eight signal dimensions in this analysis at once (research impact, adoption curves, team composition, domain lineage, and more), built on a multi-year research partnership with the University of Oxford and applied identically to every company we see. The taste matrix in this post is a small public sample of that discipline: revealed preference, reconstructed from data, instead of narrative written after the fact.

For founders, the practical difference is simple. At the six firms above, you need to match the persona their gate was built for. At Vela, if your strength is measurable, the pipeline sees it, warm introduction or not. That is what makes quant the seventh door. See our research program for how the models are built, and our live arena experiment for how we score those models against real startup outcomes.

Methodology and sourcesAnnounced investments, firm blog posts, press coverage, and founder background research from June 2024 to June 2026; roughly 70 seed and Series A rounds across the six firms. Signal scores (0-10) are estimated weights derived from pattern frequency across observed checks, not stated investment criteria. A gate means absence disqualified in the observed data.

Frequently Asked Questions

What do top VC firms look for in founders in 2026?

It depends entirely on the firm. Across roughly 70 seed and Series A rounds from June 2024 to June 2026, each top firm treats a different signal as its hard gate: Sequoia gates on elite research credentials, Benchmark on market adoption and revenue, a16z on top-AI-lab origin and open-source community, Founders Fund on competitive programming medals or Palantir-tier operator experience, Thrive Capital on having built at scale inside a top tech company, and Lightspeed on a prior acquisition in the exact domain. There is no universal checklist. There are six different ones.

How do I know which VC firm is the best fit for my profile?

Match yourself to a founder persona, then pitch the firm whose gate you already pass. If you are a published researcher with no revenue, that is Sequoia or a16z. If you have $1M+ ARR and no pedigree, that is Benchmark. If a developer community runs your open-source project, that is a16z. If you won IOI or IMO medals, that is Founders Fund. If you built a well-known product inside Meta, Google, or Stripe, that is Thrive. If you already sold a company in your domain, that is Lightspeed. Pitching a firm whose gate you do not pass wastes months.

Does Sequoia Capital require founders to have a PhD?

A PhD is near-mandatory for Sequoia's research-track seed bets. In the 2024-2026 window, Sequoia backed tenured CMU professors (Skild AI), the AlphaChip co-creators from Google DeepMind (Ricursive), and Gemini lead researcher Misha Laskin (Reflection AI). Sequoia will write a first check with zero product and zero revenue if the researcher is singular enough. Exceptions exist when a founder has lived the domain problem firsthand, but elite research caliber verified by institutional affiliation is the primary filter.

Does Benchmark invest in pre-revenue startups?

Rarely. Benchmark is the only tier-1 fund in the analyzed set that generally waits for the market to speak first: nearly every 2024-2026 seed or Series A had $1M+ ARR or mass developer adoption before Benchmark engaged. HeyGen was profitable and at roughly $35M ARR before its check, Mercor's 21-year-old founders had $1M ARR fully self-funded, and LangChain had 84K GitHub stars. For Benchmark, adoption proof substitutes for founder pedigree.

What does a16z look for in AI founders?

Top-AI-lab origin and pre-existing open-source communities. a16z backed SSI (founded by former OpenAI Chief Scientist Ilya Sutskever, a $1B seed at zero revenue), Black Forest Labs (the Stable Diffusion co-creators, with FLUX ranked #1 on Hugging Face), and Inferact (the vLLM maintainers, whose project ran on 400K GPUs with 2,000+ contributors). For a16z, an existing community or lab credential is the proof of concept. Revenue is not required.

Does Founders Fund invest in first-time founders?

Yes, but through a very specific credential system. Founders Fund backed Cognition's founders at six months old with zero revenue and zero prior exits because the 10-person team collectively held 10 IOI (International Olympiad in Informatics) gold medals, with CEO Scott Wu ranked #1 globally in 2014. It is the only firm in the set where competition medals function as a stand-alone qualifying gate. A parallel channel runs through the Palantir and Anduril operator ecosystem for executive-led teams.

What kind of founders does Thrive Capital back?

Builders at scale inside top tech companies who left to solve a problem they lived through. Chai Discovery's Joshua Meier built ESM1, the first protein language transformer, at Meta before founding Chai; ElevenLabs' Piotr Dabkowski published at NeurIPS and shipped ML at Google. Thrive writes roughly 12 checks per year across all stages and uses seed bets as relationship-incubation vehicles for later $1B+ growth checks. Thrive's founder pairs typically knew each other 5 to 15 years before founding.

What does Lightspeed look for in seed-stage founders?

A prior acquisition or research impact in the exact domain the founder is now applying AI to. Granola's Chris Pedregal had two Google acquisitions (Socratic, Stack) before his seed round; Skild AI's founders are two tenured CMU professors with a decade of Meta FAIR robotics research. Lightspeed's vertical AI thesis is organized by profession, and its 11-office global structure surfaces European founders (Granola in London, Mistral in Paris, ElevenLabs in Warsaw) that US-first funds systematically miss.

Do you need a prior exit to raise from a top VC firm?

No, but you need a substitute the specific firm accepts. In the 2024-2026 data, prior exits scored highest at Lightspeed (preferred, 7/10) and were largely irrelevant at Founders Fund and Thrive (2-3/10), where IOI medals and top-tech building track records substitute. Sequoia substitutes research credentials, Benchmark substitutes revenue, and a16z substitutes lab origin and open-source adoption. The practical question is not whether you have an exit. It is which firm's gate you already pass.

What is a tier-1 quant VC and how does it evaluate founders?

A quant VC applies quantitative models, AI agents, and reproducible research to sourcing and investment decisions instead of relying on one partner's pattern matching. Vela Partners is the leading tier-1 quant VC: it encodes founder signals like the ones in this analysis (research impact, adoption curves, team composition, domain lineage) into scoring models built with its University of Oxford research partnership, and applies them consistently across every deal. For founders, the practical difference is that a quant VC evaluates the full signal profile rather than one hard gate, so strength that is measurable gets seen even without a warm introduction.