search_models
Search models
Find the exact id of an LLM tracked by InferIndex from a name or partial name (e.g. 'deepseek', 'qwen3 max', 'claude opus'). Returns matching model ids and names, best match first.
Parameters
queryrequired
InferIndex
io.github.InferIndexPublished by GitHub account InferIndexv1.0.0Updated Oct 24, 2026
LLM API prices across 70+ providers: cheapest offer, comparisons, history and cost estimates.
Overview
AI-generated overview
Remote MCP server that lets an assistant compare LLM API prices across 60+ providers, check price history, estimate costs and weigh self-hosting against API use
- What you can do
- Find tracked model ids by name or partial name, with the best match first and optional filters for context, tools, JSON, vision, region, no training on prompts and open sign-up, plus usage figures for cost estimates. Compare current offers for one model, one line per provider and source, cheapest first, with quantization, conditions and reliability. Review price history per offer and the lab's official prices by days, date range or a single date. Estimate workload cost per request and per month, including tiered pricing and prompt-cache prices. List tracked GPU types with the lowest rental price per tier, and rental offers for one GPU. Judge whether self-hosting an open-weights model on rent
- When to use it
- Use it when choosing a provider or model for a workload, checking whether a quoted price is competitive, tracking how prices moved, or deciding between self-hosting open-weights models and paying for an API.
- Requirements
- A remote Streamable HTTP endpoint; no installation, account or key is required. An optional API key can be sent in the x-api-key header to raise rate limits. Network access to the service is needed; requests are limited per IP and per API route.
Before you install
Read-only and unauthenticated, but prices are collected from public listings and parsed pages, so treat them as what the collector last saw, not as a billing guarantee; confirm on the provider's own page before committing to real spend. Promotions, misparsed units and reseller margins can distort figures. The API is early, has no versioned contract, and response shapes may change.
Installation
In SourceWeft
- Open InferIndex in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.
Use with an agent
Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.
Read https://sourceweft.com/mcp/io.github.InferIndex%2Finferindex/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.Other MCP clients
Add this to your client's mcpServers config.
{
"mcpServers": {
"inferindex": {
"type": "http",
"url": "https://mcp.inferindex.dev/mcp"
}
}
}README
InferIndex
InferIndex finds the cheapest provider for a given LLM, across more than 60 tracked pricing sources — direct provider APIs and aggregators/routers — with price history and promotion detection.
Live API: https://api.inferindex.dev
Service status: https://status.inferindex.dev (external uptime monitoring)
This repository holds the public documentation, examples and (eventually) a client SDK for that API. The backend itself — collection, parsers, data model — lives in a separate private repository.
Example response as of 2026-10-07 — prices, providers and statuses change.
Prices are in USD per million tokens, converted at the day's ECB rate when the source publishes another
currency. blended_per_1M is (3 × input + output) / 4, a rough proxy for a typical chat workload.
Each offer appears once, at its cheapest source: via is "direct" or the aggregator it comes from, and
also_via lists other sources selling the same offer. promo is false unless a promotion was detected, in
which case confidence and price_before_promo are filled in.
Status
Early / actively developed. The API is public and free to use, without authentication, but there is no uptime guarantee, no versioned API contract yet, and endpoints or response shapes may change. If you build something on it, expect to adjust when they do. Feedback and issues are welcome.
Service status: https://status.inferindex.dev (external uptime monitoring). It shows current and past availability; it is not an uptime guarantee.
Why
Provider pricing pages disagree, change without notice, and rarely show what resellers actually charge for the same model. InferIndex collects prices from more than 60 sources, each read from its own listing on its own schedule (from hourly to daily depending on the source), stores every change, and answers "who is cheapest for model X right now" from that history — client requests never trigger a live call to a provider.
Data freshness and limits
- Cache: successful responses carry
Cache-Control: public, max-age=300(5 minutes) (except/health/ready, never cached). On the custom domain, responses are additionally held in Cloudflare's edge cache for the same 5 minutes. - Collection frequency: most sources are polled hourly; a few (exchange rates, model catalog) once a day.
- Rate limits: 60 requests per minute per IP on
/cheapest,/history,/resellers,/self-host,/gpusand/gpu-rentals, counting only requests not answered from cache. An optional API key (x-api-keyheader) raises this to 300 requests per minute per key; free keys will be offered, self-service sign-up isn't available yet. An invalid key returns 401. Over the limit, requests get429. The MCP server has its own limits, see Use with AI assistants. - Cost estimate: add
prompt_tokens,output_tokens,cached_ratioandrequests_per_dayto/cheapestor/resellersto get an estimated cost per request and per month for each offer, andsort=estimated_costto rank by it. See docs/api.md. - Hidden tiers by default: degraded service tiers (
flex,batch— lower priority, variable latency, in exchange for a lower price) are excluded from/cheapestand/resellersunless you ask for them withinclude_tiers=flex,batch. The number hidden is reported inhidden_tiers.
Reliability of the prices — read this before trusting a number
Data corrections are listed in the changelog.
InferIndex is a comparison tool, not a source of truth. Treat every price as "what our collector last saw", not as a guarantee of what you will be billed:
- Prices come from public APIs and, for some sources, HTML pages parsed with source-specific code — a page redesign can silently break a parser until it's caught.
- A price that looks unusually low may be a promotion, a misparsed unit, or a provider that hasn't updated its page yet. InferIndex applies guard rails (rejecting implausible prices, quarantining sudden jumps, flagging outliers against a second reference source) and reports detected promotions with a confidence level, but these are heuristics, not certainties.
- Reseller aggregators can add their own margin. When a source's pricing model isn't fully understood, it is either excluded or flagged.
- Always confirm the price on the provider's own page or dashboard before committing to real spend.
Endpoints
/cheapest filters
sort=blended\|input\|output— sort order (defaultblended)min_context=100000— minimum context windowquantization=fp8,bf16— accepted quantizationsmin_uptime=95— minimum 30-minute uptime (%), as observed by an aggregator (not measured by InferIndex)tools=true,json=true,vision=true— only offers declared to support tool calling, JSON output, image inputregion=eu— only offers processed in that region (eu, us, cn, uk, ch, sg, id, my, vn, kr, jp, in, ca, au), per the providers' official documentsno_training=true— only offers whose providers state they don't train on promptsno_waitlist=true— only providers whose signup is open to everyonestrict=true— also drop offers for which the filtered data is unknown (by default they're kept and flagged inunverified, counted infilters_unknown)include_tiers=flex,batch— include degraded service tiers (excluded by default, see above)explain=true— list every excluded offer with its reason codes (anexplanationsummary is always included)
Offers also carry conditions (data region, retention, training on prompts, and what it takes to open an
account — each with its official source link), quantization_source (whether the compute precision is declared
by the source or inferred), and, on /resellers, reliability from the providers' official status pages. See
docs/api.md.
/history
granularity=raw returns one row per price change; day / week (weeks start Monday) return one point per
period and per offer (provider, quantization, variant) with min, max and last price. Use days or from/to
for a period, or at=2026-09-15 for the prices in force at a given moment. Prices are converted to USD at the
ECB rate of the day of the price.
Model search (/models):
Example response as of 2026-09-15 — prices, providers and statuses change.
/health/ready
Example response as of 2026-09-15 — prices, providers and statuses change.
status is ok, degraded (a non-critical part is having trouble — see the degraded array for which) or
unavailable (no usable price data). HTTP 200 for ok/degraded, 503 for unavailable. Useful as an
uptime-monitor target if you depend on this API.
Full route and parameter reference: docs/api.md.
Use with AI assistants (MCP)
InferIndex is also available as a remote MCP server, so an AI assistant can look up prices for you:
- Endpoint:
https://mcp.inferindex.dev/mcp(Streamable HTTP) (https://api.inferindex.dev/mcpalso works, as an alias) - Listed in the official MCP Registry as
io.github.InferIndex/inferindex. - Read-only, no authentication. An API key can be passed in the
x-api-keyheader. - Limits: 120 requests per minute per IP on
/mcp, at most 10 JSON-RPC messages per batch and 64 KB per request (otherwise400or413with JSON-RPC error-32600). Over the rate limit:429with JSON-RPC error-32000. Each tool call also counts toward the limit of the API route it uses (/cheapest,/resellers,/history,/self-host,/gpus,/gpu-rentals).
Every result includes api_url, the equivalent API call, so you can check or reuse it.
cheapest, compare_providers and estimate_cost return a compact view by default: the winner in detail and one short
line per other offer. Pass detail: "full" to get every field of every offer. In the compact view, an absent condition
was not published by the provider (it never means "no"), and a flag that is not listed in signals is false.
Setup
For Cline and other agents that install servers themselves, see llms-install.md.
Claude Code
With an API key, add --header "x-api-key: YOUR_KEY".
Claude Desktop / claude.ai: Settings → Connectors → Add custom connector, URL https://mcp.inferindex.dev/mcp.
Older Claude Desktop versions, in claude_desktop_config.json:
Cursor: one click, or add it by hand.
In ~/.cursor/mcp.json:
With an API key, add "headers": { "x-api-key": "YOUR_KEY" } next to "url".
Codex CLI
Or in ~/.codex/config.toml:
With an API key, add env_http_headers = { "x-api-key" = "INFERINDEX_API_KEY" } to that table and set the
INFERINDEX_API_KEY environment variable.
Contributing
Documentation fixes and example contributions are welcome — open a pull request or an issue. This repository covers the public API surface only; the backend itself is not open source.
License
- Documentation (this README,
docs/): CC BY 4.0. Reuse and adapt it freely, with credit to InferIndex. - Code (examples, snippets, client SDK): MIT.
These licenses cover this repository only, not the InferIndex backend or its price data.
Source: README.md at commit 0344797
Tools
8cheapest
Cheapest offers for a model
Cheapest current API offers for one model across direct providers and aggregators, in USD per 1M tokens (input, output, blended 3:1). Stale prices, and flex/batch tiers, are excluded by default. Optional filters (context, tools, JSON, vision, region, no training on prompts, open sign-up) and usage (tokens per request, requests per day) to get an estimated cost per request and per month. Returns the winner in detail and one short line per following offer. A condition that is absent was not published by the provider, it never means "no"; a flag not listed in signals is false.
Parameters
jsonlimitmodelrequiredtoolsdetailregionstrictvisionmin_contextno_trainingno_waitlistcached_ratio+4 morecompare_providers
Compare providers for a model
Current offers for one model, one line per provider and source (direct or via an aggregator), cheapest first (10 by default), with price, quantization, signals and training on prompts when published; the cheapest offer comes in detail. Pass detail: "full" for context, every published condition (data regions, sign-up) and reliability from official status pages on each offer.
Parameters
sortlimitmodelrequireddetailregionstrictno_trainingno_waitlistcached_ratioinclude_tiersoutput_tokensprompt_tokens+1 moreprice_history
Price history of a model
Price history of one model: every offer tracked by InferIndex (daily or weekly min/max/last price in USD, or raw price changes), plus the official price of the model's lab over time. Give either days, or from/to (YYYY-MM-DD), or at (a date) for the prices in effect that day.
Parameters
attodaysfromlimitmodelrequiredprovidergranularityestimate_cost
Estimate the cost of a workload
Estimated cost of a workload on one model at each provider: cost per request, and per month if requests_per_day is given, taking the provider's tiered pricing and prompt-cache price into account. Offers sorted by estimated cost, cheapest first: the cheapest in detail, then one short line per offer with its estimated cost.
Parameters
limitmodelrequireddetailcached_ratiooutput_tokensprompt_tokensrequests_per_daylist_gpus
GPU types tracked for rental prices
GPU types whose rental price InferIndex tracks (H100, A100, L40S…), with the lowest price per GPU per hour in USD for each tier (guaranteed, community, spot) and the number of GPUs of that configuration. Prices are per GPU per hour, as published by each provider, and dated. Tiers are different products and are never compared with each other.
Takes no parameters.
gpu_rentals
Rental offers for one GPU, by tier
Rental offers for one GPU type: provider, price in USD per GPU per hour, number of GPUs in the published configuration, billing and region when published. One block per tier (guaranteed, community, spot), each with its own cheapest offer, also given per configuration (cheapest_by_gpu_count: the price per GPU of a single GPU and of a multi-GPU node are not comparable); tiers are never compared with each other. guaranteed: capacity the provider presents as not interrupted; community: third-party hosts; spot: interruptible capacity. Prices are as published by the provider, and dated. Get the GPU id from list_gpus.
Parameters
gpurequiredtierdetailregiongpu_countself_host_or_api
Self-host an open model or use the API?
Is it cheaper to host an open-weights model yourself on rented GPUs, or to use the cheapest API offer? Returns a one-sentence verdict (headline) and the numbers behind it: the break-even volume in million tokens per day, the throughput the whole configuration must deliver for self-hosting to cost less (required_tokens_per_second: compare it with what your setup does), the break-even utilization when a throughput is known, the GPU configuration and its hourly price, and what the verdict rests on (confidence). Without a volume you still get the verdict and the break-even volume. It is an estimate: the throughput is a published measurement (measured), an estimate scaled from one (derived; estimated for a wide range), or a published minimum (lower_bound: a floor, valid for requests of up to 2,048 tokens in total, that can show self-hosting wins, never that the API is cheaper); or you give your own tokens_per_second. When no published figure decides, the headline gives the break-even volume and the required throughput, to compare with yours, rather than a yes or no, and may point out a dearer setup that published figures do decide (settled_alternative: for comparison, not a recommendation). headline_kind and headline_values give the headline as fields. It declines to give a number (status refused, with a reason) when it cannot do so reliably, for example a closed model, weights that do not fit the requested GPU, or a GPU nobody rents. Hardware rental only: the hourly price the provider publishes for the machine. It does not count the engineering to deploy and run the model, monitoring, redundancy, start-up time, or storage and network costs billed separately.
Parameters
gputiermodelrequireddetailgpu_countutilizationquantizationoutput_tokensprompt_tokenstokens_per_dayweights_marginrequests_per_day+1 moreVersion history
1- v1.0.0LatestSep 18, 2026
Score
100% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 daysRequired25 pts
- Auth declaredRequired10 pts
- Tools listedRequired15 pts
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts