Tools, costs, and commitments

Research date: September 14, 2026. All prices below were checked against official pricing pages or recent (mostly past-12-months) secondary sources on or shortly before this date. Prices, free tiers, and startup programs change frequently — sometimes monthly. Several tools covered here repriced within the last year (GitHub Copilot Pro+ nearly doubled in July 2026; Replit cut its Core price in February 2026; Lovable overhauled its credit system in August 2026). Always confirm on the vendor's own pricing page before budgeting. Where we could not verify a figure directly, it is marked (estimate).

A note on sources: many "X vs Y" comparison pages are published by one of the vendors being compared (e.g., Attio's "best CRM for startups," Dover's ATS comparisons, Render's Railway-vs-Fly page, Pilot's Bench FAQ, Gusto's Rippling comparison). We cite some of these for factual pricing claims but flag the conflict of interest wherever a vendor is grading its own homework. Aggregator and affiliate review sites are also common in this space; we prefer official pricing pages and treat third-party roundups as corroboration, not gospel.

How to read this chapter

For each category we compare the major options on: current price, free tier and its cliffs, ease of use, startup discounts, scalability, lock-in/data ownership, and best-fit stage. At the end, we assemble eight concrete tool stacks with estimated monthly costs.

Three cross-cutting truths worth internalizing before the tables:

1. "Free" tiers are marketing funnels with cliffs. Some cliffs are gentle (Postgres row limits you can export around); some are sharp (Supabase pausing your database after a week of inactivity, Lovable pausing your production app's auth and database when credits run out, Amplitude quoting $1,500+/month the moment you outgrow the free tier). We flag the sharp ones throughout and summarize them near the end.
2. Startup credit programs can cover most of your first two years of infrastructure — but they expire, usually in 12 months, and the card on file gets billed the day they do. Architect as if you were paying.
3. Lock-in is mostly about data gravity, not contracts. Anything storing your system of record (cap table, CRM, accounting ledger, analytics history, email list) is expensive to leave. Anything stateless (hosting, CI, design tools) is cheap to leave. Weight your diligence accordingly.

Startup programs: the free-money layer

Before comparing individual tools, know the major programs, because they change the effective price of everything else.

Sources: Startup Cloud Credits 2026: Tiers, Rules, Cliffs (Causo Hub), Microsoft Learn: Changes to the Microsoft for Startups program (July 2025), The Register on Azure credit changes, HubSpot for Startups, Notion for Startups, Stripe Atlas, Analytics free-tier comparison 2026 (AgentDeals).

Practical guidance: raise a priced round or sign with a partnered accelerator before applying to the big-credit tiers — the delta between the self-serve tier ($1k–$5k) and the investor-linked tier ($100k+) is the largest "discount" available to any startup. And don't multi-cloud just to stack credits; the operational cost of running two clouds usually exceeds the credit value.

1. Company formation

All five services below file the same Delaware (or Wyoming) paperwork; you are paying for workflow quality, document quality, and what happens after formation.

Verified pricing: Stripe Atlas ($500 + $100/yr, $2,500 credits, 83(b), SAFEs); the Atlas/Firstbase/doola comparison figures from Kanopy Labs' 2026 incorporation comparison; Clerky's ~$427/$819 package structure via SparkLaunch's Clerky pricing page and Rho's Clerky review (Rho is a banking vendor — conflict noted). LegalZoom figures are estimates; its pricing page churns constantly (CheckThat LegalZoom pricing).

The honest summary: if you are raising venture capital, the choice is Stripe Atlas (fastest, best perks) or Clerky (most lawyer-proof paperwork); the ~$100 difference is noise. If you are a non-US founder building a bootstrapped business, doola's LLC path costs less over three years than a Delaware C-corp's franchise-tax-and-agent overhead. Don't incorporate a C-corp you don't need — Delaware franchise tax plus registered agent plus a required corporate tax filing is real annual drag (budget roughly $400–$800+/yr even at zero revenue, plus tax prep).

2. Legal documents

This category is unusual: the best tools are free, and paying more mostly buys workflow, not better contracts.

• YC SAFE documents — the standard for early fundraising. The post-money SAFE templates (cap, discount, MFN, plus side letter and pro rata) are free from Y Combinator, and Cooley GO's free YC SAFE generator will assemble a signature-ready set for US, UK, and Singapore companies. Stripe Atlas and Clerky both generate SAFEs inside their platforms too. Never pay to generate a standard SAFE.
• Cooley GO — free incorporation packages, board consents, NDAs, offer letters, and financing docs from a top venture law firm (Cooley GO documents). Conflict of interest is benign but real: it is a client-acquisition funnel for Cooley.
• Common Paper — free, open-source standard agreements (Cloud Service Agreement, DPA, NDA, Design Partner Agreement, and a growing SaaS toolkit) with a freemium contract-management platform on top (Common Paper standards, Technical.ly on the SaaS toolkit). For a B2B SaaS startup, adopting Common Paper's CSA instead of paying a firm to draft bespoke terms can save $5,000–$15,000 in year-one legal fees (estimate based on typical firm rates).

When to actually pay a lawyer: priced equity rounds, anything with IP assignment ambiguity (university spinouts, ex-employer issues), regulated products, and M&A. Template-stacking through those situations is how startups buy expensive problems.

3. Domains and websites

Domain pricing is a solved problem with one right answer and many expensive ones.

Source: NameBuddy 2026 registrar guide. GoDaddy is omitted deliberately: aggressive renewal pricing and upsells make it strictly dominated for startups.

Recommendation: register at Cloudflare (or Porkbun if you want registrar-level flexibility), always pay for multi-year on your primary domain, and treat .ai's ~$70–$90/yr as a marketing expense, not a necessity. For the website itself, see Framer/Vercel/Netlify below — a launch-stage marketing site should cost $0–$30/month.

4. Product design and prototyping

Verified against Figma's pricing page.

The seat model is the thing to understand. Figma bills per editor; anyone who only views, comments on, or inspects a file is free on a paid plan. A 20-person company with three designers and two front-end engineers who actually edit pays for five seats, not twenty. Audit editor seats quarterly — they accumulate silently when someone opens a file, nudges one component, and gets auto-upgraded.

Free-tier cliff (moderate): 3 projects and 2 members is tight, and the third teammate triggers the upgrade. At $12/editor/month this is among the least painful cliffs in this chapter.

Lock-in (high, universally accepted): Figma files are a proprietary format. PNG/SVG/PDF export works fine; exporting a live design system into another tool does not, realistically. Everyone tolerates this because the alternatives are worse.

On AI design features generally. Figma Make, Framer's AI site generation, and the prompt-to-app tools in §5 now overlap heavily. The honest read: they are very good at producing a plausible first draft and exploring visual directions fast, and they are not yet a substitute for someone who understands your users. Treat the output as a starting point, not a deliverable.

5. Software development

Verified against GitHub pricing, Cursor pricing, Replit pricing, and CloudZero's Claude pricing breakdown.

Three things that matter more than the price list

Cost is now usage-shaped, not seat-shaped. Cursor, Claude Code, and Replit all bundle "a set amount of model usage" into the subscription and meter overage on top. This makes budgeting genuinely harder than classic per-seat SaaS: five engineers on $20 plans is not reliably a $100/month line item, because heavy agentic use pushes into on-demand billing. Set spend limits on day one. Counterintuitively, for teams doing sustained agentic work the $100–200/month Max-style plans are frequently cheaper than metered API billing for equivalent volume — price both before committing.

Prompt-to-app platforms and AI-native editors are different products, and founders pick wrong. Replit and Lovable generate and host a whole application from prompts, aimed at people who are not primarily engineers. Cursor and Claude Code accelerate engineers working in an existing codebase. If you have a production repo, you want the editor category. If you are validating an idea over a weekend and do not intend to maintain the result, the generator category is fine — with the real caveat that these codebases are often awkward to hand to an engineer later.

Read the IP and training terms. Check each vendor's stance on training against your code and whether zero-data-retention is available. Most now ship privacy modes (Cursor advertises team-wide privacy mode on Teams). If enterprise customers will eventually security-review you, having this documented in advance saves a painful scramble.

Free-tier cliff (soft): GitHub's free tier is genuinely generous, and Actions overage is pay-as-you-go rather than a wall. The real escalation is Enterprise (from $21/user/mo) for SAML SSO and advanced security — which is exactly what enterprise customers start demanding. Lock-in is low: git clone gets your code. Issues, Actions workflows, and Projects are not portable, and that is the actual switching cost.

6. AI development platforms

Foundation model API pricing (per million tokens)

Verified against Anthropic's API pricing docs and OpenAI's API pricing page.

The four discounts that actually move the bill

1. Batch processing: 50% off input and output on both Anthropic and OpenAI. If your workload tolerates latency — nightly enrichment, eval runs, bulk classification — this halves cost for roughly zero engineering effort beyond a queue.
2. Prompt caching. Anthropic charges 1.25x base input for a 5-minute cache write and 2x for a 1-hour write, but cache reads are 0.1x base input. For agents carrying large stable system prompts or documents, this is usually the single largest lever available.
3. Model routing. The spread from Haiku ($1/$5) to Fable ($10/$50) is 10x. Most production tasks do not need the top tier. Route by task, gated on evals.
4. Output length discipline. Output costs 5x input across Anthropic's lineup, so verbose responses are the expensive failure mode.

Data residency (US-only) carries a 1.1x multiplier on Anthropic; OpenAI lists +10%. This matters for the regulated stack in §21.

Orchestration and open source

• Vercel AI SDK — free, open-source TypeScript toolkit abstracting across providers with streaming and tool-calling built in. The practical benefit is provider-swappability: changing models becomes a config change instead of a rewrite. Cheap insurance against both repricing and model deprecation.
• Hugging Face — free for public models and datasets; Pro ~$9/mo, Team ~$20/user/mo, Inference Endpoints billed by compute hour (estimates — verify current). The hub is the de facto registry for open-weight models.
• Self-hosting open models (Llama, Mistral, Qwen and successors) via Together, Fireworks, Groq, or your own GPUs. The economics rarely beat API pricing until you have sustained, high, predictable volume, because you pay for idle GPU time. It wins decisively when you need data isolation, custom fine-tunes, or availability independent of a vendor.

Flag: model deprecation is a live operational risk. Providers retire versions. Pin them, keep evals in CI, and build behind an abstraction so a forced migration is configuration rather than a lost quarter.

7. Databases

Verified against Supabase pricing, Neon pricing, and PlanetScale pricing.

Supabase's inactivity pause deserves a specific warning. Free projects sleeping after one week is the sharpest cliff in this chapter for side projects and demos you show intermittently — your database goes to sleep and the demo breaks in front of the investor. The two-active-project cap also bites earlier than expected. At $25/month, Pro removes both and is the correct purchase the moment anything is customer-facing. The larger strategic issue is the $25 → $599 gap to Team: if enterprise customers require SOC 2 attestation from subprocessors, your floor is $599/month.

Neon's model is the most startup-friendly here because paid tiers carry no monthly minimum — you pay for compute-hours consumed, and scale-to-zero pushes idle cost toward nothing. Database branching (a branch per pull request, with production-like data) genuinely changes how teams test migrations. The cost is forecastability: CU-hour billing is harder to predict than a flat fee. Set alerts.

On lock-in. Neon, Supabase, and PlanetScale Postgres are all Postgres — schema, queries, and pg_dump output are portable, and migrating between them is a weekend, not a quarter. Supabase adds proprietary surface (Auth, Storage, Edge Functions, Realtime) that is not portable; the more of that platform you adopt beyond the database, the higher your exit cost. Firebase is the outlier: its document model, security rules, and SDK coupling make migration a data-layer rewrite. That is a defensible trade for a consumer app needing offline sync and realtime out of the box — just price it deliberately rather than discovering it in year three.

8. Hosting and cloud

Verified against Vercel pricing, Render pricing, Fly.io pricing, and Cloudflare plans.

Hetzner is roughly an order of magnitude cheaper than the hyperscalers for equivalent raw compute, and the included traffic is the real story: 20TB on EU plans versus hyperscaler egress that can dominate a bill. EU catalog as of September 2026: CX23 at €5.49/mo (2 vCPU / 4GB / 40GB NVMe), CX33 at €8.49/mo (4 vCPU / 8GB). The US catalog is priced less generously on traffic — CPX11 at $20.49/mo with 1TB (secondary source; Hetzner's own page uses a configurator). Note the older CX22/32/42/52 line was replaced by CX23/33/43/53 at higher prices, so pre-2026 Hetzner price comparisons are stale. What you give up: managed services, a serious IAM system, and conveniences you would otherwise not build. Excellent for a technically strong team that wants predictable costs and will own operations.

Cloudflare R2 charging no egress fees is the structural pricing story of the last few years. Storage at $0.015/GB-month, Class A operations $4.50/million, Class B $0.36/million, and no data-transfer-out charge — versus S3 egress that routinely becomes a top-three line item for media-heavy or inference-heavy workloads. If you serve large objects at volume, this single decision moves your infrastructure bill more than anything else in this chapter.

Vercel's Hobby tier is explicitly non-commercial, and it is the most-violated term in startup infrastructure. If you are charging money, you need Pro at $20/month. Pro includes a $20 usage credit then meters: Edge Requests $0.60/million beyond 1M, Function Invocations $0.60/million beyond 1M, Fluid Active CPU $0.128/hour beyond 4 hours, build minutes $0.007–$0.105/minute by machine type. The flat-rate CDN with spike protection addresses the old fear of a viral post producing a shocking invoice, but configure spend management anyway.

Render's free instances spin down when idle, producing cold starts — fine for a demo, not for a product. Its $25 → $499 Pro-to-Scale gap is a hard cliff, and HIPAA-compliant options plus SAML SSO sit on the far side, which drives the regulated stack in §21.

Fly.io has no plan minimum. Shared-cpu machines from about $1.94/month, with a 40% discount for annual reservations (e.g. $36/year for $5/month of usage). Volumes $0.15/GB-month. Egress $0.02/GB in North America and Europe, $0.04/GB in APAC and South America, $0.12/GB in Africa and India; inbound is free. The best option for genuinely multi-region applications at small scale.

9. Payments

Verified against Stripe's pricing page and Paddle's pricing page.

The merchant-of-record decision

This is the most consequential choice in this section, and it is not primarily about price.

As a processor, Stripe moves money; you remain merchant of record. That means you are legally responsible for collecting and remitting sales tax, VAT, and GST everywhere you have obligations. For a US company selling to US customers, manageable. The moment you sell digital products to consumers in the EU, UK, or a few dozen other jurisdictions, you acquire registration and filing obligations that scale with the number of jurisdictions, not with revenue. Stripe Tax computes and in some configurations files — but the liability stays yours.

A merchant of record (Paddle, Stripe Managed Payments) becomes the legal seller. They absorb tax liability, handle global remittance, defend chargebacks, and appear on the customer's statement. You are effectively selling to them; they resell to the end customer.

The price of that transfer: Paddle at 5% + 50¢, or Stripe Managed Payments at 3.5% on top of standard Payments fees — roughly 6.4% + 30¢ all-in on a domestic card — against Stripe's bare 2.9% + 30¢. You are paying roughly 200–350 basis points for global tax compliance.

Worth it when: you sell digital products or SaaS to consumers and small businesses globally, you have no finance staff, and the alternative is VAT registration in a dozen countries. Paddle's claim that the à la carte equivalent "typically totals ~7% and above" is self-serving vendor framing, but directionally fair — tax software, chargeback management, and dunning each cost money and attention.

Not worth it when: you sell primarily domestically; your volume makes 200–350bp a serious number ($3,500–$5,000/month at $100k MRR); or you are B2B enterprise with invoicing and reverse-charge VAT, which is far simpler than B2C.

The Lemon Squeezy situation — verified

Stripe acquired Lemon Squeezy in July 2024 to build "a global merchant of record solution" (Lemon Squeezy's own announcement; corroborated by Silicon Republic).

As of the company's January 2026 update, the platform remains operational and is still accepting signups, with no announced sunset date and no forced migration timeline. But CEO JR Farr has been explicit about direction: "Our goal is to provide Lemon Squeezy users an easy way to migrate to Stripe Managed Payments," with public (currently invite-gated) access to Managed Payments expanding across 35+ countries.

Practical read: do not start a new business on Lemon Squeezy in late 2026. It works today and there is no deadline, but you would be building on a platform whose owner has publicly described the migration path away from it. Start on Stripe Managed Payments (if you want MoR inside the Stripe ecosystem) or Paddle (if you want an independent MoR that is not owned by your processor). Existing Lemon Squeezy merchants are not in danger and should plan migration on their own schedule rather than urgently.

Flag — concentration risk. With Lemon Squeezy absorbed, the independent MoR market for small software companies is meaningfully thinner. Paddle is now the primary independent alternative at scale. Weigh that if vendor diversity matters to you.

10. Banking and treasury

Verified against Mercury's pricing page.

The post-SVB consideration persists. The March 2023 Silicon Valley Bank failure permanently changed how startups think about deposits. Standard FDIC insurance is $250,000 per depositor, per insured bank, per ownership category — trivially exceeded by a company that just closed a seed round. Mercury and competitors address this via sweep networks distributing deposits across partner banks to multiply effective coverage. Confirm the current sweep arrangement and stated coverage directly with the provider, because these programs change and the details are the entire product.

Two rules that survived 2023: hold operating cash at more than one institution, and know how long payroll survives if one provider is unavailable for a week.

Mercury is free for core banking with no minimums, making it the default. Plus at $29.90/month adds invoicing with ACH debit, recurring invoices, and unlimited 1099 filings — worth it only if you actually invoice. Pro at $299/month buys a relationship manager and suits larger companies. Treasury access gates at a $250,000+ balance, which is the relevant threshold for post-raise companies wanting yield on idle cash.

Brex and Ramp are primarily corporate cards and spend management, not banks, and both monetize substantially through interchange — which is why the core products are free. Ramp's positioning is cost control (it actively flags duplicate subscriptions and negotiates vendor pricing); Brex's is depth for venture-backed companies with complex expense workflows. The practical differentiator for most startups is which integrates more cleanly with your accounting system.

Underwriting note: both underwrite on cash balances rather than credit history, which is why a pre-revenue startup with $2M in the bank gets a real limit. The corollary is uncomfortable — your limit falls as your runway shortens, precisely when you would most want it.

Flag: Ramp and Rho both publish extensive comparison content about this category while competing in it. Treat "best startup bank" listicles on those domains as marketing.

11. Accounting and bookkeeping

Verified against QuickBooks pricing and Puzzle's pricing page.

Bench: what happened, and where it stands

On December 27, 2024, Bench Accounting abruptly ceased operations, emailing customers that the platform would be "immediately inaccessible" — with no advance warning to roughly 650 employees, many of whom lost their jobs without severance. Bankruptcy filings in Canada on January 7, 2025 disclosed $2.8 million in cash against $65.4 million in liabilities (Wikipedia; Accounting Today).

On December 30, 2024 — three days later — Employer.com announced it had acquired Bench. Bench resumed operations in January 2025 as an Employer.com subsidiary, still under the Bench brand, with headquarters moved from Vancouver to San Francisco.

Current status (September 2026): Bench is operating under Employer.com ownership.

The lesson is not "avoid Bench." It is that your books are load-bearing data and you must hold your own copy. Customers with only Bench's proprietary platform faced losing years of financial records overnight, days before tax season, with no warning. Regardless of who does your bookkeeping:

1. Export your general ledger, trial balance, and supporting documents monthly to storage you control.
2. Prefer providers that work inside a standard system you own. Pilot and most bookkeeping firms operate within your QuickBooks or Xero account — if the relationship ends, you keep the ledger. This is materially safer than a proprietary platform.
3. Bank and processor statements are the ultimate backstop. They can always be re-downloaded and books rebuilt from them. Painful, survivable.

(Flag: Pilot published an explainer on the Bench situation and Acuity published another. Both are direct competitors to Bench and their framing should be read accordingly — the underlying facts above are corroborated by Accounting Today and bankruptcy filings.)

Choosing

Pre-revenue, few transactions: a spreadsheet plus disciplined receipt storage is defensible for a few months. Do not let it run past your first tax filing.

Bootstrapped with revenue: QuickBooks Simple Start ($38/mo) or Xero. Xero's unlimited-users-on-every-plan model is a real advantage once a bookkeeper, an accountant, and a founder all need access — QuickBooks caps users per tier. Add a part-time bookkeeper at $300–800/month rather than doing it yourself; founder hours are the expensive input.

Venture-backed: Puzzle ($60–100/mo) is built for startups: accrual books by default, with burn, runway, and margin surfaced natively — that runway number is what your board asks about monthly. Pilot (~$600–700+/mo) buys actual accountants and is right when the alternative is a part-time controller.

Flag: accrual accounting matters more than founders expect. Cash-basis books make SaaS revenue look erratic and will not survive investor diligence. Set up accrual from the start; converting later is expensive.

12. Payroll and HR

Verified against Gusto's pricing page and Deel's pricing page.

Gusto is the default for a US-only startup under ~50 people, because payroll is a compliance function whose failure mode is IRS penalties. It files federal, state, and local taxes, handles W-2s and 1099s, and is simple enough that a founder can run payroll untrained. Ten employees on Simple is $49 + $60 = $109/month. Note the tier trap: multi-state payroll requires Plus at $80 + $12/person — hire your second employee in a different state and you have changed tiers.

Rippling's pitch is consolidation: payroll, benefits, device management, app provisioning, and identity in one system, so onboarding provisions a laptop and accounts in a single action. Genuinely valuable at 30+ people. Below that, the integration benefit is smaller than the added complexity, and modular pricing makes the total harder to forecast.

Deel dominates international hiring, and its pricing explains why: EOR at $599/employee/month is what it costs for a third party to legally employ someone where you have no entity. That is usually far cheaper than establishing a foreign subsidiary (five to six figures plus permanent compliance). Contractor management at $49/contractor/month is the lighter path most distributed startups use first.

The expensive mistake: misclassifying employees as contractors. Rules differ by country and, in the US, by state — California's ABC test is materially stricter than the federal standard. Penalties include back taxes, benefits, and interest. Deel's Contractor of Record at $325/month exists precisely because this risk is real enough that companies pay to transfer it. If someone works full-time hours, on your schedule, with your equipment, under your direction, they are probably an employee in most jurisdictions regardless of what the contract says.

Justworks is a PEO: it co-employs your staff, letting a five-person startup access benefits pricing normally reserved for much larger employers. That pooled-benefits advantage is the whole reason to choose a PEO. The tradeoff is less flexibility and an awkward exit — unwinding co-employment means standing up your own benefits and state registrations.

13. CRM and sales

Verified against Attio's pricing page and HubSpot's CRM pricing page.

The overwhelmingly common mistake is buying a CRM too early. For your first 20–50 customers, a spreadsheet or a Notion database tracks deals perfectly well at zero cost. A CRM earns its place when more than one person is selling, when deals run long enough that you forget state, or when a board wants pipeline reporting.

HubSpot's Starter at $7/seat/month annual is genuinely cheap, and the free tier (2 users, 1,000 contacts) covers a founding team. The caution is escalation: Professional and Enterprise cost multiples of Starter, and the model is explicitly to make entry painless and growth expensive. The HubSpot for Startups program advertises up to 90% off year one, stepping to 50% then 30% — but it is widely reported that the 90% tier requires being in a partnered accelerator or VC portfolio, and that most applicants receive 30% (elefante RevOps analysis — secondary source; HubSpot does not publish tier criteria). Model year two and year three at list price before committing, because the step-down is where the pain lands.

Attio is the current favorite among technically-minded startups. Its free tier is unusually generous (3 seats, 50,000 records), and the differentiator is a flexible data model — you define your own objects instead of accepting a fixed Lead/Contact/Opportunity schema. That suits non-standard motions: PLG, community-led, investor relations. At $35/seat/month annual it costs more per seat than HubSpot Starter; the flexibility is what you are buying.

Salesforce is correct when you have a large sales org, complex territory and approval rules, and budget for an administrator. Under 50 people it is almost always over-specified, and true cost includes implementation consultants, not just seats.

Lock-in (medium): records export as CSV, so the data is portable. Workflows, automations, custom fields, reports, integrations, and your team's habits are not. Switching CRMs mid-scale is among the more painful migrations in this chapter. Choose once, deliberately, at the point you genuinely need one.

14. Marketing and email

Verified against Resend pricing, Loops pricing, and beehiiv pricing.

Separate transactional from marketing email — by subdomain, and ideally by vendor. Transactional mail (password resets, receipts, notifications) must arrive; marketing mail is allowed to land in Promotions. Mixing them on one sending domain means a bad campaign's spam complaints can degrade deliverability for your password resets. This is a five-minute architectural decision with multi-year consequences.

Resend is the default transactional choice for developer teams: clean API, React Email for templating, 3,000/month free. Pro at $20/month for 50,000 emails with $0.90/1,000 overage is simple to reason about. The free tier's 100/day cap binds well before the monthly limit — a signup spike hits it first.

Loops prices on subscribed contacts, not sends, with unlimited sending on paid plans. Favorable if you run high-frequency lifecycle email to a modest list; unfavorable if you hold a huge list you email rarely.

beehiiv's free tier to 2,500 subscribers is the most generous in the newsletter category, and it includes the monetization and ad-network features rather than gating them. Scale at $43/month and Max at $96/month both cover up to 100,000 subscribers — excellent per-subscriber economics near the top of that band, mediocre just past 2,500.

Free-tier cliff (sharp, category-wide): subscriber-count escalation defines this category. Mailchimp is particularly known for steep growth curves and for counting unsubscribed contacts in some plan calculations — prune your list. Model your cost at 10,000, 50,000, and 100,000 contacts before choosing, because the rank order of these vendors changes substantially across those thresholds.

Deliverability is a discipline, not a feature. Configure SPF, DKIM, and DMARC; warm new sending domains gradually; monitor bounce and complaint rates; drop hard bounces immediately. No provider rescues you from sending to a purchased list.

15. Analytics

Verified against PostHog's pricing page and Plausible's pricing.

PostHog's free tier is the standout offer in this entire chapter. 1M events, 5,000 session replays, and 1M feature-flag requests per month — per product, resetting monthly — with usage stopping at the limit rather than generating a surprise invoice unless you add a payment method. PostHog states that 97% of its companies stay within the free tier. For most seed-stage startups this genuinely costs nothing while replacing three or four separate tools. The consolidation is the real value: one event stream means going from "conversion dropped" to watching the failing session recording is one click, not a data-joining project.

The tradeoff is breadth over depth. Amplitude and Mixpanel have more sophisticated behavioral cohorting and retention analysis, and analytics-heavy growth teams often prefer them at scale.

Plausible at $9–19/month (to 10k pageviews) answers a different question: simple, privacy-respecting web traffic analytics, no cookie banner requirement, a script small enough not to affect page performance, no personal data collected. It does not do product analytics. Running Plausible on the marketing site and PostHog in the product is a sensible and common split.

GA4 is free and universal and therefore the default — but be clear-eyed: it is complex, its data model confuses people arriving from Universal Analytics, the free tier samples large queries, and it is a Google advertising product first. If you run paid acquisition on Google, you need it. Otherwise it is optional.

Free-tier cliff (sharp, at the top end): Amplitude is the notable one — outgrowing its free tier commonly triggers enterprise quotes in the $1,500+/month range (estimate), with little in between. Know where that edge sits before you build dashboards your team depends on.

Data ownership matters more here than almost anywhere, because your event history is the record of how the product actually performed. PostHog and Plausible are open-source and self-hostable, which puts a genuine floor under your options. GA4, Amplitude, and Mixpanel offer export, but the raw event stream is not straightforwardly yours. Pipe events to your own warehouse in parallel from the start — cheap insurance, and it becomes the foundation for analysis these tools cannot do.

16. Customer support

Verified against Intercom's pricing page.

Start with a shared inbox. For the first hundred customers, support@yourcompany.com routed to a shared Gmail or Google Group is adequate and free. Buy a helpdesk when multiple people answer, when you need SLA tracking, or when threads start getting lost.

The outcome-based AI pricing shift is the significant 2026 development. Intercom's Fin bills from $0.99 per resolution rather than per seat, and works standalone alongside other helpdesks including Salesforce. This is a structurally different cost model: you pay for tickets actually deflected, not for capacity. For a startup with unpredictable support spikes that can be favorable — but model it carefully. At 5,000 resolutions/month, $0.99 each is $4,950, which exceeds the seat cost of a small support team. The economics favor AI resolution when volume is high, questions are repetitive, and the alternative is hiring.

Crisp's workspace-based pricing (rather than per-seat) makes it notably cheap for the common startup pattern where several people occasionally answer tickets and per-seat pricing is punitive.

Plain targets technical B2B products: API-first, native Slack and Linear integration, designed for engineering teams who handle support in tools they already live in rather than context-switching into a helpdesk.

Lock-in (medium): conversation-history export quality varies and matters more than founders expect — it is your record of what customers actually complained about. Check export capability before accumulating three years of tickets.

17. Project management and docs

Verified against Linear pricing, Notion pricing, and Slack pricing.

Slack's 90-day free-tier history limit is the sharpest data cliff in this chapter. Your team's decision history, context, and institutional memory silently become inaccessible after three months. Teams that stay on free Slack for a year discover they have no record of why anything was decided. Either pay for Pro ($7.25/user/month annual) or write decisions down somewhere permanent — the second is better practice regardless, because chat is a poor system of record.

Linear's 250-issue free tier arrives faster than expected — an active team of three can hit it in a couple of months. At $10/user/month the fix is cheap. Linear is the strong default for software teams, and the reason is opinionation: fast, keyboard-driven, and it enforces a workflow rather than offering infinite configuration. Teams wanting to model an unusual process find it constraining; most teams benefit from the constraint.

Notion's free-tier cliff is structured around team size specifically. An individual gets effectively unlimited blocks; the moment the workspace has 2+ members, block creation becomes limited. This reliably catches founders who build an extensive wiki solo and then add a cofounder. Plus at $10/member/month resolves it.

A composition note: Linear + Notion + Slack is close to a standard for venture-backed software startups. At 10 people on annual billing that is roughly $100 + $100 + $72.50 ≈ $272/month. Cheaper alternatives exist; the integration quality of this particular combination is why it persists.

18. Recruiting

Do not buy an ATS to make your first five hires. A spreadsheet, a calendar, and a structured interview rubric are sufficient and cost nothing. An ATS earns its keep when you run multiple roles concurrently, need compliance records, or have enough candidate volume that people fall through cracks.

The headcount-based pricing model is the thing to understand about Ashby. Because it charges on total company headcount rather than recruiter seats, cost scales with company size rather than recruiting activity — a 250-person company that hires slowly still pays as a 250-person company. Reported real-world ranges of $30k–$70k/year at 100–300 employees and $60k–$120k/year at 300–500 come from Truffle's pricing analysis — a direct competitor, so treat the figures as directional — though the structural point is verifiable from Ashby's own published model. Ashby's product quality, particularly its analytics, is widely regarded as best-in-class; the question is whether the pricing shape fits yours.

Flag: both Dover and Truffle publish "Ashby pricing" and "Ashby alternatives" content while competing in the category. The $400/month Foundations figure is Ashby's own published price; the higher ranges are competitor-sourced estimates.

The highest-leverage recruiting investment is not software. It is a written, structured interview process — defined competencies, consistent questions, scorecards completed before group discussion. That costs nothing and improves hiring quality more than any tool in this table.

19. Security and compliance

Tier structure verified against Vanta's pricing page — note that none of the three vendors publishes dollar figures, which is itself informative about the category.

Realistic first-year SOC 2 Type II cost

All of these are estimates, triangulated from Cavanex's 2026 SOC 2 cost breakdown and similar secondary sources. Audit costs vary widely by scope, number of trust service criteria, and firm. Get quotes.

The critical distinction founders miss: compliance software is not an audit. Vanta, Drata, and Secureframe automate evidence collection, monitor controls continuously, and manage policy attestation. They do not issue the SOC 2 report — an independent CPA firm does, as a separate engagement at separate cost. Budget for both.

Type I vs Type II. Type I attests that controls are designed appropriately at a point in time. Type II attests that they operated effectively over a period, typically 3–12 months. Enterprise buyers generally want Type II, and the observation window means the timeline cannot be compressed arbitrarily — starting a Type II three months before a deal closes does not work.

When to start. Honestly: when a deal requires it, and not before — with one exception. If you know enterprise sales is your motion, start implementing the controls early (SSO everywhere, enforced MFA, least-privilege access, encryption at rest, documented on/offboarding, vendor inventory) even without buying the platform. Retrofitting controls onto a sloppy environment is the expensive part; the software mostly watches controls that already exist.

Other frameworks: HIPAA (US healthcare), PCI-DSS (card data — largely avoidable by never touching card numbers and letting Stripe handle them), GDPR (EU personal data — a legal and process obligation more than a software purchase), ISO 27001 (international, frequently required alongside SOC 2 by European buyers).

20. Cap table and investor management

Verified against Carta's pricing page and Pulley's pricing page.

Carta's Launch tier is free up to 25 stakeholders and $1M raised, which covers a founding team plus early angels and is genuinely sufficient pre-seed. Pricing beyond that is not published — Carta states each package "has a price per stakeholder with a minimum annual fee," meaning you negotiate. The $1M-raised ceiling is the cliff, and it expires at precisely the moment you close your first meaningful round.

Pulley publishes its prices, which in this category is worth something on its own: $1,200/year for 25 stakeholders, $3,500/year for 40 stakeholders with 409A valuations included. The half-stakeholder counting for small angels (≤$50k) is a thoughtful touch for companies with long angel lists. Pulley has no free tier, so the comparison is $1,200 versus $0 at the earliest stage — and inverts once you exceed Carta's free limits.

Two things matter more than the tool.

First, your cap table must be correct. Errors — an unrecorded option grant, a SAFE whose recorded terms differ from the signed document, a missing board consent — surface during diligence and can delay or kill a financing. Whatever you use, reconcile against signed documents at least annually, and always before a raise.

Second, 409A valuations are a recurring cost founders forget to budget. You need one before issuing options, and roughly annually or after any material event such as a priced round. Carta and Pulley bundle them into higher tiers; standalone providers charge roughly $1,500–$5,000 (estimate). A defensible 409A is what protects employees from adverse tax treatment on exercise.

Lock-in (high): cap table data exports, but migrating platforms requires re-verifying every security against source documents, which is real work. This is a choose-once decision. Carta's scale and auditor familiarity is a genuine advantage heading toward an institutional round; Pulley's price transparency is a genuine advantage if you want to know what you will pay.

21. Eight recommended stacks, with estimated monthly cost

All figures below are estimates assembled from the verified prices above, on annual billing where available, and assuming no startup credits applied. Credits (see the startup programs table near the top) can drive the infrastructure lines close to zero for the first 12–24 months. Actual costs vary with usage.

(1) Solo founder, pre-revenue

Goal: spend nothing until something works.

Notes. Amortized formation aside, the only genuine cost is one AI coding subscription. The moment you charge a customer, Vercel Hobby's non-commercial term obliges a move to Pro (+$20) and you should move Supabase/Neon to a paid tier. Budget ~$50–75/month at first revenue.

(2) Small bootstrapped SaaS (2–5 people, some revenue)

Priorities: low fixed costs, avoid per-seat escalation, merchant-of-record if selling globally.

Notes. The AI coding line has the widest variance — a $200 Max plan for a two-person team doing heavy agentic work is often cheaper than metered billing. Paddle's 5% + 50¢ is the right call here specifically because a small bootstrapped team cannot absorb global VAT compliance. At $20k MRR that is ~$1,000/month, which is the price of not hiring a finance person.

(3) Venture-backed B2B software company (20–40 people, Series A)

Priorities: enterprise sales readiness, compliance, scalability. Per-seat tools and compliance dominate.

Notes. Two lines deserve scrutiny. AI coding at $800/month for 20 seats is now a standard line item and generally defensible on productivity grounds — but audit real usage, because seat sprawl here is endemic. Compliance at ~$2,000/month amortized is the price of enterprise sales eligibility; if you are not selling to enterprises, delete the line and save $25k–$70k in year one.

(4) Consumer app startup

Priorities: scale to many users cheaply, mobile infrastructure, growth analytics.

Notes. The dominant cost is not in this table. App Store and Google Play take 15–30% of in-app revenue, dwarfing every SaaS line combined — structuring for web checkout where permitted is worth more than every optimization in this chapter. Second: bandwidth and storage are the infrastructure risk for media-heavy consumer apps, which is exactly why Cloudflare R2's zero-egress model appears here. It can be the difference between a $200 and a $2,000 monthly infrastructure bill.

(5) AI startup

Priorities: model costs dominate; everything else is rounding error.

Notes. This is the one stack where tool selection barely matters and unit economics are the entire business question. Know your cost per user, per session, and per successful outcome — weekly, not quarterly. The four levers in order of impact: (1) route to the cheapest model that passes evals (the Haiku-to-Fable spread is 10x); (2) prompt caching, where Anthropic cache reads are 0.1x base input; (3) batch processing at 50% off for anything asynchronous; (4) control output length, since output costs 5x input. Apply all four before considering self-hosting, which only wins at sustained high volume or when data isolation demands it. And use the Vercel AI SDK or equivalent so a provider change is configuration rather than a rewrite — real insurance against both repricing and model deprecation.

(6) Hardware startup

Priorities: CAD, PLM, supply chain — categories largely outside this chapter and far more expensive than software tooling.

Notes. Software tooling is genuinely the small line here. Tooling, molds, certification (FCC/CE/UL), and inventory are where hardware capital goes, and those are five- to seven-figure items no SaaS decision affects. Two chapter-relevant points: QuickBooks Plus at $140/month is the minimum tier with inventory tracking (Simple Start and Essentials do not include it), and cash flow is the binding constraint because you pay for components months before collecting revenue. A corporate card with a real limit and clean PO tracking matters more here than in any software stack.

(7) Regulated startup (fintech / health)

Priorities: compliance from day one. The compliance layer is neither optional nor cheap.

Start from the venture-backed B2B stack (3), then add and modify:

Notes, and these are the important ones.

Vendor tier selection changes completely. Features you would otherwise treat as optional — HIPAA support, SOC 2 attestation from subprocessors, SAML SSO, audit logging, data residency — all sit on the expensive side of the cliffs flagged throughout this chapter. Concretely: Render's HIPAA option is on Scale at $499/month, not Pro at $25. Supabase's SOC 2 / ISO 27001 posture is on Team at $599/month, not Pro at $25. Neon's HIPAA is on Scale. Anthropic's US-only data residency carries a 1.1x multiplier; OpenAI lists +10%. Every one of these is a 5–25x jump from the tier a non-regulated startup would choose — which is why the regulated stack is not "stack (3) plus Vanta."

You need a Business Associate Agreement (health) or equivalent contractual posture (fintech) with every vendor touching regulated data. Not every vendor will sign one, and that constraint, not price, should drive selection. Verify before building, not after.

Regulatory counsel is not a tool and not optional. Fintech in particular may require money transmitter licenses, bank partnerships, or sponsor-bank relationships whose costs and timelines dwarf everything in this chapter. Talk to a specialist lawyer before writing code.

(8) Global / distributed company

Priorities: hiring and paying across borders, async work, global tax compliance.

Notes. EOR is the dominant line and it is worth it. At $599/employee/month, five international employees cost ~$36,000/year — against establishing five foreign subsidiaries at five to six figures each plus permanent compliance obligations. Use contractors ($49/month) where the relationship genuinely is contractor-shaped and EOR where it is not; misclassification is the expensive mistake, which is precisely why Deel sells Contractor of Record at $325/month to absorb that risk.

Two other global-specific points. A merchant of record is more valuable here than in any other stack, because selling from and into many jurisdictions is exactly the problem MoR solves — Paddle's 5% + 50¢ replaces VAT/GST registration across dozens of countries. And Fly.io's multi-region deployment genuinely matters when users are distributed, because latency is a product feature. Also budget GDPR compliance as an ongoing process obligation — DPAs with every subprocessor, data mapping, deletion workflows — not a tool you can buy.

22. Cross-cutting guidance

The sharpest free-tier cliffs, ranked by how badly they surprise founders

1. Slack — 90-day message history. Institutional memory silently disappears. Highest-consequence cliff in this chapter.
2. Supabase — free projects pause after 1 week of inactivity, max 2 active projects. Your demo breaks exactly when you need it.
3. Vercel Hobby — non-commercial only. Widely violated; charging money obliges Pro.
4. Supabase Pro $25 → Team $599. A 24x jump, with SOC 2 / ISO 27001 on the far side.
5. Render Pro $25 → Scale $499. HIPAA and SAML SSO on the far side.
6. Amplitude free → enterprise quote, commonly $1,500+/month (estimate) with little in between.
7. Notion — block limits activate at 2+ members. Catches solo founders who add a cofounder.
8. Linear — 250 issues. Arrives within weeks for an active team; cheap to fix at $10/user.
9. Figma — 3 projects, 2 members. The third teammate triggers it.
10. Carta Launch — 25 stakeholders and $1M raised. Expires exactly when you close your first real round.
11. Resend — 100 emails/day binds well before the 3,000/month limit.
12. Email marketing generally — subscriber-count escalation is steep and vendor rank-order changes across thresholds. Model at 10k / 50k / 100k contacts.

Data ownership, ranked

Low lock-in (a standard underneath): GitHub/GitLab (Git), Postgres-based databases (Neon, Supabase core, PlanetScale Postgres), Cloudflare R2 (S3-compatible API), PostHog and Plausible (open-source, self-hostable), SMTP-based email.

Medium lock-in: CRMs (records export; workflows and habits do not), helpdesks (check conversation export quality), Notion (export is lossy for databases), Figma (proprietary format, universally tolerated).

High lock-in: Firebase (document model + security rules + SDK coupling; migration is a data-layer rewrite), payroll providers of record (mid-year transitions are genuinely painful), cap table platforms (migration means re-verifying every security against source documents), proprietary bookkeeping platforms (see Bench, §11).

The general rule: prefer tools with a standard underneath, and keep your own copy of anything catastrophic to lose — books, event history, customer records, cap table.

Six recurring mistakes

1. Buying too early. Spreadsheets are a legitimate CRM, ATS, and project tracker for the first several months. Buy when the pain is real, not when you imagine it.
2. Optimizing for the wrong horizon. Choosing a tool because it is cheap at 3 people when you will be 30 within a year — or paying enterprise prices at 3 people for scale you will not reach. Model both.
3. Treating credits as free money. They expire in 12–24 months and encourage architecture you would not otherwise choose. Architect as though you were paying.
4. Not exporting your own data. Bench's December 2024 shutdown left customers days from tax season with no access to years of records.
5. Ignoring the merchant-of-record decision until VAT is already owed. Deciding MoR posture after selling into twenty jurisdictions is far more expensive than deciding it before.
6. Seat sprawl. Per-seat tools accumulate seats for people who opened a file once. Audit quarterly — Figma editors, GitHub seats, Slack accounts for departed contractors, and AI coding seats are the usual offenders.

A closing note on AI-tool pricing volatility

Every AI-adjacent vendor in this chapter has repriced within the last twelve months, and the direction is not uniformly up: Anthropic made Claude Sonnet 5's introductory $2/$10 rate permanent and cancelled a scheduled increase to $3/$15. Simultaneously, pricing models are shifting — from per-seat to usage-based (Cursor, Claude Code, Replit) and from per-seat to per-outcome (Intercom Fin at $0.99 per resolution). Three practical implications: set spend limits on every usage-billed tool on day one, re-check pricing quarterly rather than annually, and prefer abstractions that make switching providers a configuration change.

Sources

All URLs below were accessed during the week of September 8–15, 2026. Vendor-published comparisons of competitors are flagged; official pricing pages are treated as authoritative for that vendor's own prices only.

Company formation and legal

• Stripe Atlas
• Clerky pricing
• Firstbase pricing
• Sparklaunch — Clerky comparison (affiliate/aggregator)
• Kanopy Labs — Stripe Atlas vs Firstbase vs doola (aggregator)
• Rho — Clerky review (competitor-published)
• CheckThat — LegalZoom pricing (aggregator)
• Y Combinator SAFE documents
• Cooley GO — YC SAFE generator
• Cooley GO — SAFE financing documents
• Common Paper standards
• Technical.ly — Common Paper SaaS toolkit
• NameBuddy — 2026 registrar guide (aggregator)

Startup credit programs

• AWS Activate credits
• Google for Startups Cloud Program
• Microsoft for Startups — program changes
• The Register — Microsoft Azure startup program changes
• Causo — startup cloud credits compared 2026 (aggregator)
• HubSpot for Startups
• Notion for Startups

Design and software development

• Figma pricing
• GitHub pricing
• Cursor pricing
• Replit pricing

AI development platforms

• Anthropic API pricing
• OpenAI API pricing
• CloudZero — Claude plan and API pricing (secondary)

Databases, hosting and cloud

• Supabase pricing
• Neon pricing
• PlanetScale pricing
• Vercel pricing
• Render pricing
• Fly.io pricing
• Cloudflare plans
• Hetzner Cloud
• Hetzner 2026 plan pricing and benchmarks (secondary/affiliate)
• Hetzner price adjustment notice

Payments

• Stripe pricing
• Paddle pricing
• Lemon Squeezy — Stripe acquisition announcement
• Lemon Squeezy — 2026 update on Stripe Managed Payments
• Silicon Republic — Stripe acquires Lemon Squeezy

Banking, accounting, payroll

• Mercury pricing
• QuickBooks Online pricing
• Puzzle pricing
• Bench Accounting — Wikipedia (shutdown, bankruptcy filings, acquisition)
• Accounting Today — Bench acquired by Employer.com after sudden shutdown
• Pilot — Bench FAQ (competitor-published)
• Acuity — Bench Accounting shuts down (competitor-published)
• Gusto pricing
• Deel pricing
• Ramp — Mercury alternatives (competitor-published)

CRM, marketing, analytics, support

• HubSpot CRM pricing
• Attio pricing
• elefante RevOps — HubSpot startup discount tiers (secondary)
• Resend pricing
• Loops pricing
• beehiiv pricing
• PostHog pricing
• Plausible pricing
• AgentDeals — analytics free tier comparison 2026 (aggregator)
• Intercom pricing

Project management, recruiting, compliance, cap table

• Linear pricing
• Notion pricing
• Slack pricing
• Truffle — Ashby pricing analysis (competitor-published)
• Dover — Ashby ATS review and alternatives (competitor-published)
• Vanta pricing
• Cavanex — SOC 2 compliance cost 2026 (secondary)
• Carta pricing
• Pulley pricing

Compiled September 2026. All pricing subject to change — verify on vendor pages before purchasing or budgeting.