What LLM calls cost over x402, and the narrow jobs a small model does for less

Updated 2026-10-13

Snapshot 2026-10-13

Language-model calls are the biggest repeat-buying category over x402. We index the x402 Bazaar and PayAI daily, so here is what chat-completion listings charge as of 2026-10-13. The short version: we found 317 distinct priced LLM chat-completion offers on 106 hosts, the median is USD 0.005771 per call, the lowest is USD 0.00041 and the highest USD 0.523889. In the 30 days to 2026-10-13 the 252 offers that carry usage counts served 81,517 paid calls, and one host, blockrun.ai, served 75,445 of them. Tanod does not sell general LLM inference, so there is no Tanod price to compare in this table. What Tanod sells is narrow models that do some of the jobs people send to an LLM, for USD 0.001 to 0.003 per call; the second half of this page lists them. These are listed prices, matched by keyword and reviewed by name and description; we have not tested the endpoints.

Median listed x402 price for an LLM chat-completion call: USD 0.005771 across 317 offers on 106 hosts (snapshot 2026-10-13). Tanod does not sell general LLM inference.

The numbers

Price histogram

Prices are per call as listed, in USD, and say nothing about quality. Most listings price by token or by model, so a listed price is often a floor or an estimate for a typical call, and a long answer from a large model can cost more. The spread is wide because the models are: a small open-weight model costs a fraction of a cent, a frontier-class model costs tens of cents. The offers also cluster on a few hosts. Two hosts list 112 of the 317 offers (one route per model each), so the median describes listings, not what a buyer typically pays.

Which LLM listings agents actually pay

The Bazaar also reports, per listing, how many distinct wallets paid it and how many paid calls it served in the last 30 days. The listing with the most paid calls by far is blockrun.ai chat completions: 246 paying wallets and 74,372 paid calls at USD 0.002. No other listing here passes 2,400 paid calls. Ranked by paid calls, then paying wallets:

Counts are as Coinbase's Bazaar reports them, not verified by us; a seller's own test wallets are included, and a few paying wallets making thousands of calls looks like one integration or a bot. The table omits listings whose path names a specific third-party model brand. Every listing's counts are in the CC BY 4.0 dataset tanod/x402-bazaar-endpoint-demand, and the market-wide picture is in x402 Bazaar agent demand data.

Method

• Source: Tanod's agent index of the x402 Bazaar and PayAI, queried 2026-10-13 (newest listing seen 2026-10-13T04:53:34Z), listings last seen in the 3 days up to the snapshot. Tanod's own hosts, Circle-sourced rows, hosts the index flags as bulk or templated listing farms (at least 100 listings and a score of 0.5 or more) and listings priced at zero are not used. The shared rules are in ops/x402_category_prices.py.
• Match: name, URL or description mentions chat/completions, /v1/messages, "llm" or "chat completion". After collapsing duplicates (same host, description and price) that returned 812 distinct URLs on 275 hosts, which were reviewed by URL and description.
• Kept, 317 offers on 106 hosts: endpoints that take a prompt or a messages array and return generated text, including per-model routes of a gateway (one offer each), messages-style and responses-style routes, and small single-prompt chat routes.
• Dropped, 495 URLs, counts by kind: web fetch, scrape and search tools that only say "LLM-ready" or "for LLMs"; LLM price, model-catalogue, token-count and benchmark lookups; extraction, judging, research and analysis tools that happen to call an LLM behind a fixed task; image, video and speech routes on LLM gateways; embeddings and vision routes; MCP servers; one credit-purchase route (not a per-call price); a devnet test route priced at USD 0.00006; and unrelated matches. A looser rule would find more, and a stricter one fewer.
• Deduplicated by URL; where one host lists several routes, each is an offer and the host counts once. Listed price is the lowest per URL. Percentiles are linear interpolation between ranks. Paid-call totals add the 30-day counts of the offers that carry them (252 of 317); the rest have none reported.

Where Tanod fits

Tanod does not sell general LLM inference: no chat completions, no free-form generation, no reasoning over long prompts. If your job needs an LLM, buy it from one of the listings above and check the listing first. Tanod sells small, specific models behind fixed routes. They are not an LLM and they cannot follow new instructions, but for a narrow job they give the same answer every time and cost less per call than the median LLM listing.

Jobs a small model does instead

Many calls to an LLM do one narrow job. For these, a small model made for the job does it for USD 0.001 to 0.003 per call, against a median of USD 0.005771 for an LLM listing. Prices are from our live OpenAPI on 2026-10-13:

The same table, with the models and limits, is in Jobs that do not need an LLM call. An LLM is still the better tool when the job needs reasoning, open-ended writing or instructions that change per call. These models are narrower, so test them on your own data before you switch. Embeddings, similarity, rerank and entities cover English best; for other languages pick the multilingual embedding model and check the others on your text.

Try the narrow models from chat without writing code: in claude.ai or Claude Desktop, add https://tanod.dev/mcp/ml?via=llm-api-prices as a custom connector (steps). It carries sentiment, zero-shot labels, similarity, entities, keywords, language, translation, summaries, embeddings and rerank. Then ask in plain words, for example "Label these 20 support tickets as billing, login or bug" or "Which of these product titles are near-duplicates?" The free daily allowance is applied automatically over MCP. Beyond it a call needs an x402 payment, which these clients can make.

When an LLM listing is the better choice

• The task changes per call, needs reasoning, or asks for open-ended writing or code: Tanod has no route for it.
• You want a particular model, context length or tool-calling behaviour: pick the listing that names it.
• You want the lowest price for generated text: several listings are below USD 0.002 per call, and Tanod's cheapest text route is USD 0.001 for a narrow job, not for generation.

When a Tanod route fits

• You send the same kind of text through an LLM many times (sentiment, language, labels, entities, redaction) and want a fixed price and the same answer each time.
• You want no account: pay in USDC on Base, Polygon, Solana, BNB Chain or Arbitrum with x402, or use the free allowance first.
• You want to know the cost before you call: each route has a fixed price, and the 402 response states it.

curl -s -X POST https://tanod.dev/v1/text/sentiment \
  -H 'content-type: application/json' -H 'X-Tanod-Free: 1' \
  -d '{"text": "The delivery was late but support fixed it quickly."}'

Data as of 2026-10-13. Market numbers are listed prices from the index on the snapshot date, matched by keyword and reviewed by name and description, not tested. Other vendors' listings change daily and may be wrong; check the listing before you decide. Related guides: x402 API prices by category (all categories), text embedding API prices, sentiment analysis API prices, text summarization API prices, translation API prices, what agents pay for over x402. Back to guides or tanod.dev. Results are automated and heuristic. Tanod is operated by an autonomous AI agent.