routerflow.art
Reported. Published by routers and providers about their own traffic, or computed from it. We sent none of these requests.window
0

Overview

Where does inference flow?reported by OpenRouter, Vercel, TrustedRouter
What it shows

One day of OpenRouter traffic. Each model's tokens are split by the apps that call it and by the providers that served it. OpenRouter is the only router that publishes the provider layer, so only its branch reaches the provider layer.

Routers

OpenRouter is the only router that publishes token counts. Vercel AI Gateway publishes each model's share and no total, so its branch is an estimate: an assumed total, from Vercel's own public statements, times the published shares. It is hatched and marked est., and the card behind that mark gives the sources. TrustedRouter publishes its own measurements of providers and no usage, so it is not drawn.

App links

Each model page lists its top 5 apps. That mix is scaled to the model's daily tokens. Everything else is "Other and unnamed apps".

Coverage

The per-model provider charts cover about a quarter of the providers-page total, so read shares, not totals. OpenRouter excludes private and zero-retention traffic at the source.

Using it

Click a model to open its provider board. Click an app, a provider or a legend entry to isolate its flow. Esc clears.

1

Apps

Which apps send the flow?fixed: 1 weekreported by OpenRouter and Vercel
Source

OpenRouter's top 20 apps for the week. An app names itself with a request header. Traffic without one is unnamed, and only OpenRouter publishes apps at all.

Coverage

Tokens of the top 20 apps divided by OpenRouter's tokens for the same week.

Categories

OpenRouter's own labels. An app with two labels is split evenly between them.

Apps with their own figures

Kilo Code and OpenCode publish model rankings from their own users. Those are whole-app figures, whichever router or key the traffic used, so they overlap with OpenRouter's count of the same apps and are never added to it.

Other populations

The ChatGPT and Claude.ai task mixes are different users under different taxonomies. They sit in a fold and are not comparable with the app categories.

Side by side

Two coding agents publish their own usage. Tokens by model lab, each bar its own 100%.

Kilo Code: 63 weeks by lab open weights taking over a coding agent
OpenCode: models, labs and countries
Task mix in other populations ChatGPT consumer messages, Claude.ai conversations
Coding agents that publish their own usage tokscale, self-reported

2

Routers

Which routers carry it, and how do their mixes differ?reported by each router
Three routers, done properly

OpenRouter publishes token counts for everything. Vercel AI Gateway publishes daily shares by model and lab, and two top-10 rankings of apps and providers. TrustedRouter publishes its own measurements of providers. Every other router we know of is in the registry below, listed and not collected. Nothing is summed across routers.

The same endpoint, two quotes

Both routers publish a listed input price for each model on each provider. Where the same model is served by the same provider on both, the two quotes sit side by side. OpenRouter's is the provider's plain endpoint where it has one; a flex or batch tier is a different product and is used only if nothing else exists. These are quotes, not bills.

Windows

OpenRouter and Vercel follow the window control. Vercel's figure for a window is the mean of its daily shares, a model missing from a day's list counting as zero that day, because Vercel publishes no totals to weight the days by.

Lab mix

One bar per router: its tokens by the lab that built the model, same labs and colours in every bar, and the same in the Apps section. TrustedRouter publishes no usage, so it has no bar.

Open weights

OpenRouter per model through its Hugging Face id. Vercel by lab.

Tokens by model lab, one bar per router. Each bar is that router's own 100%.

Side by side

Every router we know of

The registry

3

Models

Which models, from which labs, at what price?reported by OpenRouter, Vercel, OpenCode
Rows

OpenRouter's top 20 models of the week. Shares on Vercel and OpenCode are matched by model name. [low] means the model is below that router's published top list.

Price

Listed input price per million tokens, lowest to highest across the model's endpoints on OpenRouter.

Provider board

Click a row. Effective price is what OpenRouter reports customers paid after cache reads and discounts over 30 days. Cache counts are asserted by providers. The Compare tab sets them against our own probe.

Pareto line

In the price and speed chart, the line joins providers that no other provider beats on both.

Side by side

Price spread per model: the builder against the cheapest
Open weights outside the routers Ollama pulls, Hugging Face downloads
AI Horde: a volunteer network

4

Providers

Who serves it, at what price, speed and uptime?reported by OpenRouter, ranks by VercelTTFT and availability measured by TrustedRouter
Volume

OpenRouter's providers page, rolling day, all models. By dollars = tokens times the blended effective price of that model at that provider, an estimate.

Price, derived

Quote against builder: the median, across the top 20 models, of the provider's lowest listed input price against the builder's own endpoint. Bill against quote: token-weighted effective price over listed price, 30 days.

Service, reported

From OpenRouter, over the provider's endpoints on the top models: token-weighted cache hit over 30 days, and the median tokens per second, first-token time and uptime of the latest day.

Measured by TrustedRouter

First-token time and availability from TrustedRouter's own monitor regions, all models pooled, with its sample count. Not by OpenRouter, not by the providers, not by us.

Daily share by provider, top eight and everyone else.

Side by side

For one model, does a dearer provider answer faster? One dot per model: the rank correlation between its providers' price and their time to first token.

Every provider of this model price, speed, uptime, quantization

Concentration over time Herfindahl index of all providers; below 1,500 is unconcentrated by the DOJ rule of thumb

5

Hardware

What silicon serves it?fixed: today's catalogmostly not published
The gap

No router or provider publishes which chip served a request. This section shows what does exist and marks the rest [x].

Quantization

Declared per endpoint on OpenRouter, for the top 20 models. "Not declared" is the provider leaving the field empty.

Own silicon

Three providers say on their own sites that they serve on chips they design. That is their statement, linked, and not verified by anyone here.

Darkbloom

The chip of each attested machine, from the network's public provider list.

Side by side

Providers that state their own silicon their statement, not verified

6

Public and private

How much inference is attested, and by whom?derived from OpenRouter reportsverification by the registry and TrustedRouter
Attested share

OpenRouter tokens served by providers that run TEEs: Phala, Venice, Chutes, NEAR AI. An upper bound for Venice, which also runs plain endpoints. A lower bound on private inference overall, because OpenRouter excludes private traffic.

Verified or claimed

Verified attested: TrustedRouter verifies the provider's attestation, or the registry fully verifies an endpoint under the provider's own name. Claims TEE: neither does. Phala and Venice are claimed today, so every "all attested" number is an upper bound.

Registry

The awesome-private-inference registry checks each provider's endpoints daily: can the endpoint prove the request ran in a TEE, layer by layer.

TrustedRouter

"Verified confidential" is TrustedRouter's own verification. Its privacy tiers classify routes, one model on one provider.

Own meters

Chutes meters dollars per chute. Darkbloom's tokens per day come from the team's own posts. Neither is checked by anyone else.

Share of OpenRouter tokens served by attested providers.

What a TEE costs: each TEE endpoint's listed price, as a multiple of the same model's other providers. Left of 1× is cheaper.

TEE price day by day

Median multiple across TEE endpoints, one point per collection day.

Side by side

Chutes, by its own meter share of metered dollars on TEE chutes, by day
Darkbloom network
Attestation registry: every endpoint checked
Attested providers against the field availability and first-token time, measured by TrustedRouter
Privacy tier of every route TrustedRouter's catalog, and where the serving companies are based

Sources, caveats, API everything above is from public endpoints, no API keys

Overview

Four questions, one line each
1

Cache

Do providers recognize a prompt they were just sent, and charge less for it?
What we send

Every provider of every panel model gets the same prompt of about 5,500 tokens twice, three seconds apart, pinned to that provider with fallbacks off. Three sweeps a day at varying hours. The first line is unique per pair, so the first request cannot hit any cache.

What we record

For both requests: the cached tokens the provider reported, the bill, and the time to the first token.

How the numbers are made

Recognized = cached tokens reported on the second request ÷ its prompt tokens. Second-request cost = its bill ÷ the first request's bill. A cell pools every pair so far and shows a provider's lowest variant of that model.

Other routers

The same pair, same bytes, is also sent through TrustedRouter and Vercel AI Gateway once a day, pinned to a provider the same way, on the panel models they also serve. Each router's results are kept apart and never pooled with OpenRouter's. A provider that recognizes our repeat through one router and not through another is being reached differently, not necessarily behaving differently.

Read with care

A miss is not proof of anything: the repeat can land on another replica, or the cache may be written late. A ring marks a provider that reported cached tokens on the first request, on bytes it had never seen; that is unexplained until it repeats.

We send each provider the same prompt twice, a few seconds apart, and show how much of the second one it reported as already cached.

Share of the second request reported as cached. Rows are models, columns are providers.

0% 100% cached0%: nothing recognized◯ reported a cache hit on a prompt it had never seen× serves the model, but our requests failedblank: does not serve this model

Side by side: providers, pooled over the models we probed

Endpoints of
2

Price

Who undercuts whom, who is moving, and do the cheap ones still cache?
What we read

The quoted input price of every endpoint of every panel model, every five minutes, from OpenRouter's public model pages. No key, no spend.

Reference price

The model builder's own endpoint where it has one (Z.ai for GLM, Moonshot for Kimi), otherwise the median quote. 1× means "priced like the builder".

Price moves

Each line is one endpoint's quoted input price as a percentage of where we first saw it. Every change the poll has seen is a step.

What happens on a cut

A cut of 1% or more fires a cache pair at the provider that cut and at the model builder in the same moment, to see whether the cache survives the repricing.

Each model's providers, placed by how their quoted price compares with the model builder's. Colour is how much of a repeated prompt they recognized.

Quoted input price ÷ reference, one row per model. Left of 1× undercuts the builder.

dot colour: 0% 100% of a repeat recognizedrecognized nothing│ the model builder

Prices that moved since the poll began, as % of the first price seen. One panel per model, shared time axis.

All price moves
3

Routing

Whom does the router pick, and does it move a conversation?
What we send

The first turn of the coding session with no provider named, a fresh session each time, so the router chooses. Also variants sorted by price and by throughput, and five-turn sessions with and without a session_id (the router's sticky-routing key).

What we record

Which provider served each request, its price at that moment, the cached tokens and the bill.

How the numbers are made

r is the exponent in pick probability ∝ price−r, fitted by maximum likelihood with a 95% profile-likelihood interval. OpenRouter documents 2. "Turns moved" counts consecutive turns of one conversation served by different providers.

4

Quality

Do providers give the right answer to checkable tasks?
What we send

Tasks with an answer a script can check: an invoice total (one number), a lookup in a ledger, a 300-word release note with required facts and a required last sentence, and the five turns of the coding session.

How it is graded

No judge model. Exact number, required strings, word count, valid JSON with the right keys. Temperature 0, the same bytes for every provider.

Why it matters

Every provider claims to serve the same model. If one fails checks the others pass, it is serving something else: a heavier quantization, a truncated context, a broken template.

5

What we send

Every probe, grouped by the kind of taskthe prompts in full six dimensions per probe
6

Studies

Two questions with their own pages
Reported against measured, for the same provider and model. The two quantities differ, so a large gap is a reason to look again, not a verdict.
1

Cache, by provider and model

Does an endpoint that reports a high cache-hit rate recognize our repeat?
Reported (x)

OpenRouter's cache-hit rate for that provider and model over the last 30 days: the share of all customers' prompt tokens the provider reported as cached. It depends on what customers send, not only on the provider.

Measured (y)

Our probe: the share of an identical prompt, sent again three seconds later, that the provider reported as cached. All pairs so far, pooled. Where a provider runs several variants of a model, the lowest.

How to read it

High reported and low measured: the provider's cache did not recognize an immediate repeat from us, although it reports plenty of hits overall. Low reported and high measured: its customers rarely repeat context. Neither is proof of anything at this sample size.

One dot per provider and model. Below the dashed line, it reports more than we measured.

Side by side: every pair in both sources

2

Cache, by provider

Which providers report more than we measure, across their models?
Pooling

For each provider, the mean of its reported cache-hit rates and the mean of our measured recognition, over the models where both exist. Unweighted: each model counts once.

The figure

One line per provider from what it reports (hollow) to what we measured (filled). A long line to the left is a provider that reports much more than we saw. Providers are ordered by that gap.

Read with care

A provider with one model in both sources is one observation. The table gives the number of models and of pairs behind every row.

Reported (hollow) to measured (filled), one line per provider, largest gap first.

Side by side: providers

What we hold. Every table, board and source, counted from the files at build time: rows, size, columns, first and last period, last sync. Private rows are counted and marked, never served.
0

Overview

What do we hold, and how far back does it go?counted from the tables at build time
What is counted

Every table the build writes, with its rows, bytes on disk and columns, and every board of every source with its rows. Tables are counted whole; a board's bytes are its share of its source's files by rows.

Periods

First and last period are the earliest and latest day a row is about. Last sync is the newest collection. Tables with no period column are undated and rebuilt whole every build.

Public and private

A source that states no license, or forbids redistribution, is collected on the same schedule but not served. Its rows are counted here and marked private.

The strip

One row per market table and per leaderboard source. The bar runs from its first period to its last; the tick is its last sync. Click a row to open it below.

How far back each source and table goes, oldest first.

1

Tables

Which tables exist, what does each hold, and how big is it?from the overview and the data dictionary
Rows

One row per table: the market tables (what routers, providers and our probe publish, one Parquet file each) and the leaderboard tables (board_entry, its dimensions, entity and alias tables).

Columns

Rows, size and column count are read from the files. First and last period come from the table's own date column. Tier is the provenance of its rows: reported, measured by others, measured by us, derived.

Detail

Click a table for every column with its type, unit, tier, key and description, and a DuckDB query that reads it over HTTP.

Side by side

2

Boards

Which leaderboards do we read, and what does each measure?from board, board_metric and board_coverage
Rows

One row per board of every source, public or not. Rows are the board's rows in board_entry: one per period, entity and metric, for each value it has had.

Period

How the board is dated: by day, week or month, a rolling window, as of a day upstream, or none, where the snapshot is the only time.

Detail

Click a board for its metrics with units and direction, where it lives, the license and citation that apply, and how many of its rows map to a canonical id.

Side by side

3

Sources

Who publishes it, under which license?from source and the boards
Rows

One row per source: who runs it, what kind of site it is, the provenance tier of its rows, its license and whether its rows may be redistributed.

Held

Boards, rows, first and last period and last sync are summed and spanned over the source's boards, public and private.

Detail

Click a source for its homepage, terms, the citation line it asks for, how we read it and every board of it.

Side by side

4

How to use it

How do I read a table, and what may I do with it?

What people write. Research papers and X posts about routers, gateways, inference prices, fees, caching, speed and whether endpoints serve what they claim. Collected by search, labelled by a small model, kept when they carry information. Each summary is one line written by that model and can be wrong: open the post on X to read what was said.
0

Overview

How much is there, and on what?counted from the feed at build time

Reading api/v2/feed.json…

1

Papers

Which research bears on routing, pricing and serving?arXiv and a literature search per topic
Where they come from

Daily searches of arXiv abstracts, arXiv links in the posts below, and a one-time literature search per topic that followed references and citations two hops out from known papers.

What is kept

A small model reads each title and abstract and scores how much it bears on this market. Papers a topic search marked as essential reading are kept and marked core.

Summaries

The one-line summary is the model's. The abstract is the authors'.

Reading api/v2/feed.json…

2

Posts

What are builders and buyers saying about routers and prices?X search, labelled by us
Where they come from

X searches naming routers and gateways together with price, fee, credit, cache and billing terms, run by month and by week from 2023.

What is kept

A small model scores each post from 0 to 1 for the information it carries: numbers, measurements, outages, launches and policy changes score high; praise, questions, promotion and crypto score low. Posts at 0.7 or above are shown.

What is shown

The summary is ours, written by the model, never the post's words. Likes, reposts and replies are as of collection. Open a post to read it on X; the text is not copied here.

Reading api/v2/feed.json…

3

This month

What got the most attention in the last 30 days?ranked by likes, reposts and replies

Reading api/v2/feed.json…

4

How it is made

How do I get these as data?api/v2

Reading api/v2/feed.json…