TON RPC Providers in 2026: an Honest Comparison
Every comparison page written by a vendor has the same problem: the vendor wins every row. This one is written by TONNode, and we'd rather you trust the page than click the buy button today, so let's set the frame honestly. We are the newcomer on this list. toncenter has been the default TON RPC provider since before most of the ecosystem existed, tonapi.io powers Tonkeeper, and dton has been answering GraphQL queries for years. Below is what each provider actually offers as of July 2026, where each one is the right choice, and where we think dedicated TON liteserver hosting — our thing — is worth paying for.
Prices and limits below were checked in July 2026 against public pricing pages. If you're reading this later, verify before deciding.
The comparison table
| Provider | Model | Price floor | RPS ceiling | Archive | Mempool | MCP | SLA |
|---|---|---|---|---|---|---|---|
| toncenter | HTTP API over shared liteservers | $0 (paid from ~30 TON/mo) | 1 anonymous, higher on paid | Indexed history (v3 API) | No | No | No |
| tonapi.io | Indexed REST + shared liteservers | $9.90/mo | 1,000 (REST) / 50 (shared liteserver) | Yes (indexed) | SSE event stream | No | Paid tiers |
| dton.io | GraphQL indexer | $0.02/req PAYG, plans from $32/mo | 400 | Yes | No | Yes | Not published |
| Chainstack | Multi-chain node platform, request units | $0 | Plan-dependent (RU quota) | From $49/mo | No | No | Paid plans |
| GetBlock | Shared nodes, compute-unit billing | $0 (40k req/day) | 60 (free-tier cap) | Dedicated plans | No | No | Paid plans |
| Orbs TON Access | Decentralized public gateway | $0 | Best effort | No | No | No | No |
| TONNode | Dedicated liteserver hosting | $20/mo, PAYG from $0.01/req | 400 (plans), 500 burst (PAYG) | Yes | Yes (stream) | Yes (npm) | 99.9–99.95% written |
toncenter: the default, and still the right first choice
toncenter is where every TON project starts, and that's fine. It's free, it mirrors the API shape every SDK expects, and the v3 API gives you indexed transaction history without running anything. The anonymous limit is 1 request per second — enough for a script, not enough for a bot with users. Paid tiers exist, but the billing runs through a Telegram bot and is priced in TON, starting around 30 TON per month. That's charming or alarming depending on your accounting department.
Where toncenter wins: prototypes, hackathons, CI, anything where $0 and familiarity beat guarantees. There is no SLA and no support contract, and that's the honest trade — you're using shared community infrastructure. Most searches for toncenter alternatives start the day a project outgrows exactly this.
tonapi.io: the widest indexed REST surface
tonapi.io (the Tonkeeper team) is two products. The REST API is the most complete indexed view of TON available over HTTP: parsed jetton transfers, NFT events, transaction traces, human-readable event descriptions. Tiers run from $9.90 to $890 per month, covering 1 to 1,000 requests per second. If your app needs "show me this wallet's activity as a human would describe it," nobody else matches this breadth today, including us.
The second product is shared liteservers: $50/mo for 10 RPS, $200/mo for 50 RPS, metered on a sliding window. Exceed it and you get their proprietary error 228 — undocumented in the ADNL standard, well documented in the scar tissue of TON developers. Shared means shared: your neighbors' load shapes your latency.
Where tonapi wins: indexed application data over REST. Where it doesn't: dedicated throughput, mempool-level access on ADNL, and predictable latency under someone else's burst.
dton.io: GraphQL maturity and the first MCP
dton is the analyst's choice. It's a full archive indexer queried through GraphQL, which means arbitrary questions — joins, filters, aggregations over the whole chain history — without pulling raw blocks yourself. Pay-as-you-go is $0.02 per request; subscriptions run $32/mo for 10 RPS, $128 for 20, $160 for 60, and $800 for 400. dton also shipped its own MCP server, which deserves credit: they saw the AI-agent access pattern early.
Where dton wins: complex historical queries and GraphQL ergonomics that have had years to mature. Where it doesn't: it's an indexer, not a liteserver — you're querying their database, not the network. No mempool, no raw ADNL, and $0.02 per request adds up fast for chatty workloads.
Chainstack: the multi-chain platform play
Chainstack added TON to a platform that already serves a few dozen chains. Entry is $0, archive access starts at $49/mo — genuinely cheap for full history — and billing is in request units. If your team already runs Ethereum or Solana infrastructure on Chainstack, adding TON to the same dashboard and invoice is a rational move.
Caveats: request-unit billing makes costs harder to predict than flat RPS tiers, TON is one chain among many rather than the focus, and availability from some regions (including Russia, where a large share of TON developers work) is worth verifying before you commit — sign-up and card billing may not work for you.
GetBlock: the generous free tier
GetBlock gives every registered user 40,000 requests per day at up to 60 RPS on shared TON nodes — the most generous free tier on this list by volume. Paid usage is billed in compute units after their 2025 migration away from per-request pricing. Same trade as Chainstack: multi-chain convenience, shared infrastructure, quota math instead of a simple RPS number.
Where GetBlock wins: free-tier volume for low-stakes workloads. Where it doesn't: TON-specific depth — no mempool, no ADNL-level access, no TON-native tooling.
Orbs TON Access: free and decentralized, with the matching guarantees
Orbs TON Access is a decentralized gateway run by the Orbs network: no API key, no payment, no single operator. As a piece of public infrastructure it's admirable, and for a dapp frontend that needs unauthenticated reads it's a reasonable default. But there is no SLA, no archive depth, and no one to email — by design. Treat it like the public liteserver config: a good citizen of the ecosystem, not a foundation for a business.
TONNode: what we built, and who shouldn't buy it
TONNode is dedicated TON liteserver hosting. Not a shared pool, not an indexer database — a liteserver provisioned for you, in under 60 seconds, speaking ADNL with Merkle proofs intact. Subscriptions run $20 to $999 per month (10 RPS lite up to 400 RPS archive), pay-as-you-go is $0.01 per request ($0.02 for archive queries) with bursts to 500 RPS, and terms are flexible: weekly rentals for load tests, yearly for committed workloads (−20%), and pay-per-request covers anything shorter. You get a mempool stream (pending external messages before they hit a block), priority transaction broadcast, and an MCP server installable via npm for AI agents. Dedicated and archive plans carry a written 99.9–99.95% uptime SLA with automatic service credits. Payment in GRAM or USDT — no card required, which also answers the regional-billing question above.
Now the honest part. We launched in 2026; toncenter and tonapi have years of production history we simply don't. We don't offer tonapi's parsed REST breadth — if you want jetton events as tidy JSON over HTTP, buy tonapi. We don't offer dton's GraphQL — if your workload is analytical queries over history, buy dton. What we sell is the thing none of the shared services can: single-tenant throughput, mempool access, and latency that doesn't depend on your neighbors.
How to choose
- Prototype, hackathon, CI: toncenter or Orbs TON Access. Free, zero setup, and the limits won't matter yet.
- Wallet, explorer, or app that displays chain data: tonapi.io. The indexed REST API is the most complete and will save you weeks of parsing.
- Analytics, research, historical queries: dton.io. GraphQL over full archive is the right tool, and PAYG keeps small workloads cheap.
- Multi-chain team with existing infrastructure contracts: Chainstack or GetBlock, with the regional-availability check done first.
- Trading, arbitrage, payment processing, anything latency-sensitive: a dedicated liteserver. Shared pools meter you on sliding windows precisely when the market is moving. This is what TONNode is for.
- AI agents over MCP: dton or TONNode — currently the only two on this list shipping their own MCP servers.
If the dedicated row is the one you kept re-reading, plans and PAYG details are on the pricing page, and the agent setup is described at MCP server. If it wasn't — bookmark this page for the day your retry logic becomes your main code path.