Algolia pricing trips up a lot of Next.js teams the same way: the free tier feels generous until you hit a hidden ceiling and suddenly you're looking at a non-trivial monthly bill. This post breaks down every cost lever in Algolia's current plans — records, searches, replicas, AI features — so you can budget accurately before you're locked in. How Algolia's pricing model actually works Algolia bills on three axes: records (the objects you index), search operations (API calls that query the index), and features (NLP, AI re-ranking, personalization). Unlike a flat SaaS subscription, the per-operation model means your bill scales directly with traffic and index size, which is great when you're small and painful when you go viral. Algolia's plan names changed in late 2025. The current tiers as of September 2026 are Free, Grow, Premium, and Elevate. Enterprise quoting sits on top of Elevate for very large accounts. Free tier limits: what you actually get The free tier is permanently free — no trial clock, no credit card required to start. The real limits are: 10,000 records across all indices 10,000 search requests per month 1 application (one API key namespace) Community support only No SLA AI Search features excluded 10,000 records sounds like a lot for a blog, but it isn't if you're indexing e-commerce products with variants, or if you're indexing each paragraph of a Sanity Portable Text document as its own record (a common pattern for granular highlighting). A 2,000-product Shopify store with three variants each is already 6,000 records before you add metadata fields. 10,000 searches per month is roughly 333 per day. A moderately trafficked search widget on a content site will blow past that within the first week after launch. Grow plan: the first paid tier Grow is the self-serve paid plan and the one most small teams land on. Pricing is consumption-based with a monthly minimum. Dimension Included at base price Overage rate Records 100,000 ~$0.50 per 1,000 Search operations 1,000,000/month ~$1.50 per 1,000 NLP / typo tolerance Included — AI re-ranking (NeuralSearch) Not included Add-on Replicas 20 per index Counts against record quota Team members Unlimited — Base price for Grow in 2026 is approximately $50–$60/month (Algolia quote-gates exact numbers but the public documentation and multiple indie benchmarks put it here). You pay that floor even in months with low usage. The record overage rate is the most common budget surprise. If you're indexing Sanity content with rich metadata — image URLs, category arrays, author references — a single document can expand into 3–5 records depending on how your indexing function splits it. 100,000 logical records can become 300,000 Algolia records fast. Premium and Elevate: what they add Premium is quote-only and aimed at teams with consistent high volume. Key additions: SLA with uptime guarantee (99.99% vs best-effort on Grow) Custom extension of rate limits during traffic spikes Priority support with response time commitments Advanced analytics (A/B testing, click and conversion tracking beyond 7-day retention) SSO and SAML for Algolia Dashboard access Elevate adds AI-powered ranking (NeuralSearch), Personalization, and Predict (behavioral recommendation). These are the features that actually justify Algolia over a self-hosted Meilisearch or a GROQ full-text fallback. If you're not using NeuralSearch, you're paying Algolia prices for what Typesense or Meilisearch could do at a fraction of the cost. Neither Premium nor Elevate has a published base price. Expect a minimum commitment of $500–$2,000/month for Premium contracts based on current market reports. Elevate engagements I've seen quoted are typically $3,000+/month for meaningful NeuralSearch usage. What actually triggers an upgrade from the free tier You'll hit the free limit faster than you expect if any of these apply: You're indexing Portable Text — splitting rich text into per-block records for highlight accuracy multiplies record count 5–10×. You have a search bar on a public marketing site — 10k searches/month disappears in days at modest traffic. You want query analytics — Algolia's analytics dashboard is locked on the free plan. You have no visibility into what your users are searching for. You need multiple indices — staging, production, and locale-split indices (e.g., products_en, products_de) all share the same record quota. For a Next.js + Sanity project syncing content to Algolia, I'd budget Grow from day one on any site expected to serve more than a few hundred daily active users. The free tier is genuinely useful for prototyping and internal tools. Algolia vs alternatives: cost positioning Tool Free tier Paid entry point Self-hostable Algolia 10k records, 10k searches/mo ~$50–60/mo No Typesense Cloud 10k records, 10k searches/mo ~$25/mo Yes Meilisearch Cloud 100k documents, 10k searches/mo ~$30/mo Yes Elasticsearch (Elastic Cloud) 14-day trial only ~$95/mo Yes Meilisearch and Typesense are the two I reach for when a client's search requirements don't need NeuralSearch re-ranking. Both have genuine open-source self-hosted options, which matters when you want to run search inside a Railway or Render container and pay only for compute. Algolia's advantage is the managed infrastructure and the quality of its relevance tuning out of the box — worth paying for if the team won't tune a self-hosted instance properly. What I tell clients before signing up Two things that save headaches later: Model your record count before indexing. Take 10 representative Sanity documents, run them through your transformDocumentToAlgoliaRecord() function, and count what comes out. Multiply by your full document count plus 30% for growth. If you're over 80,000, budget for Grow from launch. Turn on query analytics early. Analytics retention on Grow is 90 days. Starting on the free plan means you have no historical data when you upgrade. Get onto Grow before you need that data, not after. Algolia is a solid choice when the budget exists and the team needs managed relevance without an ML team. Just go in with your record count calculated and your search operation estimate grounded in real traffic projections — not the optimistic ones.