Route traffic to any model, any provider and region

Each request is served by a router that selects the model, provider and region. Router changes require no application changes.

Start for Free

Router types

Routing strategies for applications and AI agents.

Static Router

One model on one route. Applications call the router instead of a model, so the model, provider or prompt can change without a code change.

Fallback Router

Members are tried in order. When a provider rate limits, times out or returns a server error, the request moves to the next member.

Weighted Router

Split traffic between members by weight. Send a share of production requests to a candidate model, compare latency, errors and cost in Traffic, then adjust the weights.

Claude Proxy

Routes Claude Code through your organization. Set the platform (Anthropic, Bedrock or Vertex), the region and the allowed models, with usage recorded per person.

Codex Proxy

Run OpenAI Codex through your organization, on the organization's OpenAI key or each developer's ChatGPT plan, with usage recorded per person.

Pi Proxy

Run the Pi.Dev coding agent on each developer's own Anthropic or OpenAI account, while the organization sees every request and what it cost.

multi-router.ai Routers Prompts Traffic
Models RS
Back to Routers
Support Chatbot
enabled static productionsupport

Answers customer questions in the help widget. Every request is served by one model on one route.

rt_chatbot·Created Aug 2, 2026

Disable
Overview Traffic Activity

Model

The model that serves every request sent to this router.

anthropic · global Edit
Model
Claude Sonnet 4.6 anthropic/claude-sonnet-4.6
Context 1M · Max output 64K · Text, vision, tool use
Route
Anthropic API anthropic · standard
Forwards to api.anthropic.com.
Region
global
Provider key
Anthropic production sk-ant-…a41c
Price
$3.00 / $15.00 per 1M tokens in · out

Prompt

A versioned prompt applied to every request.

inline · latest Edit
Prompt
support-agent ip_7Hq2Lw · rev 12
You are the support assistant for {{product_name}}. Answer from the help center first…
System message
merge The caller's system message is added after the prompt.

Aliases

Other names that call this router.

Edit
Aliases
chat support
multi-router.ai Routers Prompts Traffic
Models RS
Back to Routers
Agent Backbone
enabled fallback production

Keeps the agent answering through provider outages and rate limits. Members are tried in order until one succeeds.

rt_agent·Created Jul 21, 2026

Disable
Overview Traffic Activity

Members

Member routers, tried in order. When one fails, the request moves to the next.

3 members Edit
1
Sonnet on Anthropic rt_sonnet_direct
Claude Sonnet 4.6 · anthropic · global
healthy 98.6% served
2
Sonnet on Bedrock US rt_sonnet_bedrock
Claude Sonnet 4.6 · bedrock · geo · us
healthy 1.3% served
3
GPT-5.4 rt_gpt54
GPT-5.4 · openai · global
healthy 0.1% served

Fall back on

The failures that move a request to the next member.

Edit
Retry next
429 rate limited 5xx server error Timeout after 30s Connection error
Return as is
400 bad request 401 / 403
A request the provider rejects as invalid fails the same way on every member.

Last 24 hours

Requests and the attempt that served them.

41,882 requests Edit
Attempts
1st attempt41,295
2nd attempt545
3rd attempt42
multi-router.ai Routers Prompts Traffic
Models RS
Back to Routers
Model Evaluation
enabled weighted experiment

Splits traffic between three models to compare quality and cost on real requests before switching.

rt_eval·Created Sep 30, 2026

Disable
Overview Traffic Activity

Members

Member routers. Each request goes to one of them, chosen by weight.

weights total 100 Edit
MemberWeightServed · 24h
Sonnet on Anthropicrt_sonnet_direct · Claude Sonnet 4.6 70 70.2%
GPT-5.4rt_gpt54 · GPT-5.4 20 19.7%
Gemini 3.1 Prort_gemini_pro · Gemini 3.1 Pro 10 10.1%

Stickiness

Keep each customer on the same member, so a conversation never changes model.

Edit
Sticky by
customer from the request's user field
Unsticky
Requests without a customer are split by weight on every request.
multi-router.ai Routers Prompts Traffic
Models RS
Back to Routers
Claude Code

Route: anthropic · global

enabled claude proxy engineering

Claude Code for the engineering organization. Every developer signs in with SSO; usage is recorded per person.

rt_claude_code·Created Sep 14, 2026

Disable
Overview Traffic Activity

Routing

How the model names Claude Code sends are resolved and where they are served.

Anthropic API · global Cancel Save
Route
Anthropic API anthropic · standard
Forwards to api.anthropic.com. Every servable model, one region.
13 of 13 models · 2 regions
Bedrock (Geo) bedrock · geo
Geography inference profiles. Inference stays inside one geography (US, EU, APAC, Japan or Australia).
12 of 13 models · 21 regions
Bedrock (Global) bedrock · global
Cross-region inference profiles. Inference may run in any AWS region, with no residency guarantee.
13 of 13 models · 33 regions
Bedrock (In-region) bedrock · in_region
Bare foundation-model ids. Inference stays in the single region called; the smallest model set.
7 of 13 models · 8 regions
Vertex AI vertex · standard
Vertex AI in one project and region.
13 of 13 models · 6 regions
Custom model group
Route each model with a model group. Members carry their own vendor, variant and region.
0 groups
Region
Region 2 regions · 1 with every model
globalAll 13 models
us10 of 13 models
Models
13 of 13 models routableAnthropic API · global
ServeModelVersionVendorVariantRegionWire IDPrice / 1M in · out
Claude Fable 5.1anthropic/claude-fable-5.1 5.1 anthropic standard global claude-fable-5-1 $10.00 / $50.00
Claude Opus 5.5anthropic/claude-opus-5.5 5.5 anthropic standard global claude-opus-5-5 $4.00 / $20.00
Claude Sonnet 5.5anthropic/claude-sonnet-5.5 5.5 anthropic standard global claude-sonnet-5-5 $2.00 / $10.00
Claude Opus 5anthropic/claude-opus-5 5 anthropic standard global claude-opus-5 $5.00 / $25.00
Claude Sonnet 5anthropic/claude-sonnet-5 5 anthropic standard global claude-sonnet-5 $2.00 / $10.00
Claude Opus 4.8anthropic/claude-opus-4.8 4.8 anthropic standard global claude-opus-4-8 $5.00 / $25.00
Claude Sonnet 4.6anthropic/claude-sonnet-4.6 4.6 anthropic standard global claude-sonnet-4-6 $3.00 / $15.00
Claude Haiku 4.5anthropic/claude-haiku-4.5 4.5 anthropic standard global claude-haiku-4-5 $1.00 / $5.00
multi-router.ai Routers Prompts Traffic
Models RS
Back to Routers
Codex

Route: openai · global

enabled codex proxy engineering

OpenAI Codex for the platform team. Developers sign in once; every session is recorded with its cost.

rt_codex·Created Oct 3, 2026

Disable
Overview Traffic Activity

Routing

How the model names Codex sends are resolved and where they are served.

OpenAI API · global Edit
Route
OpenAI API openai · standard
Forwards to api.openai.com with the organization’s provider key.
7 of 7 models · 1 region
ChatGPT subscription chatgpt · codex
Each developer signs in with their own ChatGPT plan. Usage is still recorded per person.
4 of 7 models
Custom model group
Route each model with a model group. Members carry their own vendor, variant and region.
0 groups
Models
7 of 7 models routableOpenAI API · global
ServeModelVendorVariantWire IDPrice / 1M in · out
GPT-5.4openai/gpt-5.4 openai standard gpt-5.4 $2.50 / $15.00
GPT-5.4 Proopenai/gpt-5.4-pro openai standard gpt-5.4-pro $30.00 / $180.00
GPT-5openai/gpt-5 openai standard gpt-5 $1.25 / $10.00
GPT-5 Miniopenai/gpt-5-mini openai standard gpt-5-mini $0.25 / $2.00
GPT-5 Nanoopenai/gpt-5-nano openai standard gpt-5-nano $0.05 / $0.40
o3openai/o3 openai standard o3 $2.00 / $8.00
o4-miniopenai/o4-mini openai standard o4-mini $1.10 / $4.40
Client endpoint
Responses APIhttps://api.multi-router.ai/v1serves every model
multi-router.ai Routers Prompts Traffic
Models RS
Back to Routers
Pi.Dev
enabled pi proxy engineering

The Pi.Dev coding agent. Each developer brings their own Anthropic or OpenAI account; the organization sees every request.

rt_pi·Created Sep 22, 2026

Disable
Overview Traffic Activity

Providers

Where Pi.Dev's requests are forwarded. Each request goes to the provider it is written for.

3 providers Edit
ProviderForwards toEndpointsCredentials
Anthropic api.anthropic.com POST /v1/messages Each person's own
OpenAI api.openai.com POST /v1/chat/completionsPOST /v1/responses Each person's own
OpenAI CodexChatGPT plan chatgpt.com POST /backend-api/codex/responses Each person's own

Usage

Requests and cost for each person, last 7 days.

12 people Edit
PersonRequestsCost
j.okafor@example.com 1,284 $42.10
m.lindqvist@example.com 962 $31.75
p.reyes@example.com 611 $18.02

More controls on every router

Traffic splits, key scopes, prompts and on/off, set per router in the dashboard or the API.

Weighted traffic splits90% Sonnet 4.6, 10% GPT-6 sol
Prompt bindingsupport-agent, pinned at rev 12
Aliasessupport, support-eu
Support rt_support
Claude Sonnet 4.690
GPT-6 sol10
Promptsupport-agent · rev 12
Aliasessupportsupport-eu
Keyscheckout-service, help-widget
API keys scoped to routersOnly checkout-service and help-widget
Enable or disableDisabled routers return router_disabled
Dashboard or APIPATCH /v1/routers/rt_support

Route to 60+ regions across 14 providers

Enterprise-grade routing that is region, latency and cost aware.

Regions and providers 63 regions · 13 APIs · select a region
ModelPublisherInput · Output
AWS Bedrock 33 regions Google Vertex 23 regions Azure 7 regions
APIs

High availability across vendors and regions

A fallback router retries a rate-limited, timed-out or failed request on its next member, on another vendor or in another region. The application receives one response.

  1. 1Anthropic serves every request.
  2. 2Anthropic returns 429 rate limited.
  3. 3The gateway automatically switches to Bedrock.
  4. 4Your application doesn't see any downtime.
Your app multi-router/agent
Multi-Router fallback router
Agent rt_agent
1Anthropic 2Bedrock
Anthropic Healthy 429
Bedrock Standby Serving
Your application receivesHTTP 200
x-mr-request-id: req_6Tn1VbQe2sx-mr-attempts: 12x-mr-provider: anthropicbedrockx-mr-region: globalus-east-1x-mr-upstream-model: claude-sonnet-4.6
Request req_6Tn1VbQe2s 200 · 1.10 s
Anthropic
Bedrock

Cost control on every request

Every request is priced from the catalog. Budgets on routers, tags or the whole organization send alerts as spend grows and block requests at the limit.

A detailed breakdown of tokens on every request

Each request records input, cache reads, cache writes by lifetime, output and tool calls, each priced at the catalog rate for its route and tier.

Priced for the exact route

Rates differ by provider, variant, region and tier. Each request is priced at the rate of the route that served it, and spend is reported by route.

Cost for every customer and every feature

Each request is tagged with your customer, the feature it came from, its session and your own request id. Multi-Router adds up cost by any of them.

Budgets for every person on a coding agent

Engineers see their spend in the portal and request more budget. Admins approve the request and see each person's limit and spend.

Request req_3Hd8LwQz5n 200
RouteOpenAI · standard · us
Modelgpt-6-sol
TierStandard
Long contextNo 61,240 of 272,000
Charged forQuantityRatePriced asCost
Input 3,420 tokens $2.20 / 1M Uncached prompt $0.007524
Cache read 52,600 tokens $0.22 / 1M 0.1× input $0.011572
Cache write · 30 min 5,220 tokens $2.75 / 1M 30 minute lifetime $0.014355
Output 1,880 tokens $11.00 / 1M 5× input $0.020680
Web search 4 calls $10.00 / 1,000 Per call, every tier $0.040000
File search 2 calls $2.50 / 1,000 Per call, every tier $0.005000
Request cost 99,131 micro-dollars $0.099131
Tokens Cost

Cache reads are 83% of the tokens and 12% of the cost. Six tool calls are 45%.

Rates for this route, per 1M tokens RATE
$2.20 in · $11.00 out · $0.22 cache read · $2.75 30m cache write
$4.40 in · $16.50 out · $0.44 cache read · $5.50 30m cache write
$1.10 in · $5.50 out · $0.11 cache read · $1.375 30m cache write
$1.10 in · $5.50 out · $0.11 cache read · $1.375 30m cache write
$4.40 in · $22.00 out · $0.44 cache read · $5.50 30m cache write
$10.00 / 1,000 web searches · $2.50 / 1,000 file search calls
MODEL
GPT-6 sol openai/gpt-6-sol RATE
OpenAI standard The vendor's ordinary endpoint for this model.
global gpt-6-sol $2.00 in · $10.00 out · $0.20 cache read · $2.50 30m cache write $4.00 in · $15.00 out · $0.40 cache read · $5.00 30m cache write $1.00 in · $5.00 out · $0.10 cache read · $1.25 30m cache write $1.00 in · $5.00 out · $0.10 cache read · $1.25 30m cache write $4.00 in · $20.00 out · $0.40 cache read · $5.00 30m cache write $10.00 / 1,000 web searches · $2.50 / 1,000 file search calls
us gpt-6-sol $2.20 in · $11.00 out · $0.22 cache read · $2.75 30m cache write $4.40 in · $16.50 out · $0.44 cache read · $5.50 30m cache write $1.10 in · $5.50 out · $0.11 cache read · $1.375 30m cache write $1.10 in · $5.50 out · $0.11 cache read · $1.375 30m cache write $4.40 in · $22.00 out · $0.44 cache read · $5.50 30m cache write $10.00 / 1,000 web searches · $2.50 / 1,000 file search calls
eu gpt-6-sol $2.20 in · $11.00 out · $0.22 cache read · $2.75 30m cache write $4.40 in · $16.50 out · $0.44 cache read · $5.50 30m cache write Not offered on this route Not offered on this route Not offered on this route $10.00 / 1,000 web searches · $2.50 / 1,000 file search calls
Bedrock global Routes to any region the vendor runs. No data residency.
eu-central-1 openai.gpt-6-sol $2.00 in · $10.00 out · $0.20 cache read · $2.50 30m cache write $4.00 in · $15.00 out · $0.40 cache read · $5.00 30m cache write Not offered on this route Not offered on this route Not offered on this route Not offered on this route
ap-northeast-1 openai.gpt-6-sol $2.00 in · $10.00 out · $0.20 cache read · $2.50 30m cache write $4.00 in · $15.00 out · $0.40 cache read · $5.00 30m cache write Not offered on this route Not offered on this route Not offered on this route Not offered on this route
Bedrock geo Inference stays within one geography.
us-east-1 us.openai.gpt-6-sol $2.20 in · $11.00 out · $0.22 cache read · $2.75 30m cache write $4.40 in · $16.50 out · $0.44 cache read · $5.50 30m cache write Not offered on this route Not offered on this route Not offered on this route Not offered on this route
ca-central-1 ca.openai.gpt-6-sol $2.20 in · $11.00 out · $0.22 cache read · $2.75 30m cache write $4.40 in · $16.50 out · $0.44 cache read · $5.50 30m cache write Not offered on this route Not offered on this route Not offered on this route Not offered on this route
Each row is one route: a vendor, variant and region with its own rates.
GPT-6 sol spend by routeThis month
RouteRequestsTokensTool feesTotal
OpenAI standard · global 412,300 $2,529.97 $197.70 $2,727.67
OpenAI standard · us 128,900 $889.39 $58.70 $948.09
OpenAI standard · eu 96,400 $666.82 $42.73 $709.55
Bedrock global · eu-central-1 71,800 $444.86 $0.00 $444.86
Bedrock geo · us-east-1 52,600 $360.35 $0.00 $360.35
Total$5,190.52
Claude Sonnet 4.6 anthropic/claude-sonnet-4.6 RATE
Anthropic standard The vendor's ordinary endpoint for this model.
global claude-sonnet-4-6 $3.00 in · $15.00 out · $0.30 cache read · $3.75 5m cache write · $6.00 1h cache write Not offered on this route $1.50 in · $7.50 out · $0.15 cache read · $1.875 5m cache write · $3.00 1h cache write
us claude-sonnet-4-6 $3.30 in · $16.50 out · $0.33 cache read · $4.125 5m cache write · $6.60 1h cache write Not offered on this route $1.65 in · $8.25 out · $0.165 cache read · $2.063 5m cache write · $3.30 1h cache write
Bedrock global Routes to any region the vendor runs. No data residency.
eu-central-1 global.anthropic.claude-sonnet-4-6 $3.00 in · $15.00 out · $0.30 cache read · $3.75 5m cache write · $6.00 1h cache write Not offered on this route $1.50 in · $7.50 out
Bedrock geo Inference stays within one geography.
eu-central-1 eu.anthropic.claude-sonnet-4-6 $3.30 in · $16.50 out · $0.33 cache read · $4.125 5m cache write · $6.60 1h cache write Not offered on this route $1.65 in · $8.25 out
ap-northeast-1 jp.anthropic.claude-sonnet-4-6 $3.30 in · $16.50 out · $0.33 cache read · $4.125 5m cache write · $6.60 1h cache write Not offered on this route $1.65 in · $8.25 out
Vertex standard The vendor's ordinary endpoint for this model.
global claude-sonnet-4-6 $3.00 in · $15.00 out · $0.30 cache read · $3.75 5m cache write · $6.00 1h cache write $6.00 in · $22.50 out · $0.60 cache read · $7.50 5m cache write · $12.00 1h cache write $1.50 in · $7.50 out · $0.15 cache read · $1.875 5m cache write · $3.00 1h cache write
eu claude-sonnet-4-6 $3.30 in · $16.50 out · $0.33 cache read · $4.13 5m cache write · $6.60 1h cache write $6.60 in · $24.75 out · $0.66 cache read · $8.25 5m cache write · $13.20 1h cache write $1.65 in · $8.25 out · $0.17 cache read · $2.06 5m cache write · $3.30 1h cache write
Each row is one route: a vendor, variant and region with its own rates.
Long contextVertex · eu · Sonnet 4.6
Prompt tokens241,000
Threshold200,000
Input rate$3.3 $6.6
Output rate$16.5 $24.75
Request cost$0.413490
At standard rates this request would be recorded at $0.219990.
Your applicationOpenAI SDK
client.chat.completions.create(  model="multi-router/replies",  user="cus_northwind",  extra_headers={    "x-mr-session-id": "ses_tk_48213",    "x-mr-correlation-id": "hd-7f3a91",  },  messages=messages,)
model
The feature's router, rt_replies
user
Your customer. Anthropic SDK: metadata.user_id
x-mr-session-id
Groups one ticket's requests
x-mr-correlation-id
Your own id for the request
Cost breakdown This month Customer›Feature›Session›Request
Customer, feature, session or requestRequestsTokensCost
▾cus_northwindcustomer 40,920$208.89
▾Reply draftsrt_replies $113.54
▾ses_tk_48213session 14115,080$0.1888
req_9Qm4TfXa2khd-7f3a91 18,220$0.013485
req_9Qm4TfXa1jhd-7f3a90 111,344$0.018615
12 more requests
▸ses_tk_48207session 973,980$0.1214
466 more sessions
▸Ticket summariesrt_summaries $61.71
▸Search answersrt_search $33.64
▸cus_globexcustomer 26,310$140.06
▸cus_initechcustomer 21,820$89.62

Employee portal

Your AI agents › Claude Code priya.n@acme.example
Your budget Monthly
$431.31of $500.0086%
Left
$68.69
Your pace
$27.60 a day
average over the last 3 days
Cap reached
in 2 days
Oct 19 at this pace
Resets
in 15 days
Nov 1, 00:00 UTC
Request limit increase
AmountPercent $750
Migrating the billing service this sprint. Needs Opus for the large refactors.
CancelSend request
Spend by day this month
Set by Acme IT
Cap
$500 per month, per person
Models
6 allowed, Opus 4.8 capped at $150
At 80%
notifies you
At 100%
requests paused until reset

Admin · Budgets

Engineering agents Routers · 4 Monthly · $500 per person
Pending requests 1
priya.n@acme.exampleClaude Code · requested 2 hours ago $500 → $750

"Migrating the billing service this sprint. Needs Opus for the large refactors."

Approve at $750 Approve Deny
Per-person limits this period
PersonOverrideEffective limitSpendStatus
priya.n@acme.example Pending $500.00 $431.31 Nearing
marcus.o@acme.example $750.00 $750.00 $611.20 Nearing
chen.w@acme.example none $500.00 $287.70 On track
ana.r@acme.example none $500.00 $500.00 Capped
Budget overrides this period: 1 approved, 1 pending. Capped requests get 429 budget_exhausted until reset.

Data residency

The gateway knows which region every route runs in. Choose a variant and region per router, or pin each model in a model group, and requests run only there.

MODEL VENDOR VENDOR VENDOR VENDOR

The vendor's ordinary endpoint for this model.

Model id
claude-sonnet-4-6
Runs in
Any region the vendor runs

Inference stays in the United States.

Model id
claude-sonnet-4-6
Runs in
United States

Inference may run in any AWS region. Cheapest, with no residency guarantee.

Model id
global.anthropic.claude-sonnet-4-6
Runs in
af-south-1, ap-east-2, ap-northeast-1, ap-northeast-2, ap-northeast-3, ap-south-1, ap-south-2, ap-southeast-1, ap-southeast-2, ap-southeast-3, ap-southeast-4, ap-southeast-5, ap-southeast-6, ap-southeast-7, ca-central-1, ca-west-1, eu-central-1, eu-central-2, eu-north-1, eu-south-1, eu-south-2, eu-west-1, eu-west-2, eu-west-3, il-central-1, me-central-1, me-south-1, mx-central-1, sa-east-1, us-east-1, us-east-2, us-west-1, us-west-2

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
eu.anthropic.claude-sonnet-4-6
Runs in
eu-central-1, eu-central-2, eu-north-1, eu-south-1, eu-south-2, eu-west-1, eu-west-2, eu-west-3

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
us.anthropic.claude-sonnet-4-6
Runs in
ca-central-1, ca-west-1, us-east-1, us-east-2, us-west-1, us-west-2

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
au.anthropic.claude-sonnet-4-6
Runs in
ap-southeast-2, ap-southeast-4, ap-southeast-6

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
jp.anthropic.claude-sonnet-4-6
Runs in
ap-northeast-1, ap-northeast-3

Inference stays in the single region the request was made to.

Model id
anthropic.claude-sonnet-4-6
Runs in
eu-west-2

Google picks the region.

Model id
claude-sonnet-4-6
Runs in
Any region the vendor runs

Inference stays in the United States.

Model id
claude-sonnet-4-6
Runs in
United States

Inference stays in the European Union.

Model id
claude-sonnet-4-6
Runs in
European Union

Inference stays in the region the router names.

Model id
claude-sonnet-4-6
Runs in
asia-southeast1, europe-west1, us-east5

The vendor's ordinary endpoint for this model.

Model id
claude-opus-4-8
Runs in
Any region the vendor runs

Inference stays in the United States.

Model id
claude-opus-4-8
Runs in
United States

Inference may run in any AWS region. Cheapest, with no residency guarantee.

Model id
global.anthropic.claude-opus-4-8
Runs in
af-south-1, ap-east-2, ap-northeast-1, ap-northeast-2, ap-northeast-3, ap-south-1, ap-south-2, ap-southeast-1, ap-southeast-2, ap-southeast-3, ap-southeast-4, ap-southeast-5, ap-southeast-6, ap-southeast-7, ca-central-1, ca-west-1, eu-central-1, eu-central-2, eu-north-1, eu-south-1, eu-south-2, eu-west-1, eu-west-2, eu-west-3, il-central-1, me-central-1, me-south-1, mx-central-1, sa-east-1, us-east-1, us-east-2, us-west-1, us-west-2

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
eu.anthropic.claude-opus-4-8
Runs in
eu-central-1, eu-central-2, eu-north-1, eu-south-1, eu-south-2, eu-west-1, eu-west-2, eu-west-3

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
us.anthropic.claude-opus-4-8
Runs in
ca-central-1, ca-west-1, us-east-1, us-east-2, us-west-1, us-west-2

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
jp.anthropic.claude-opus-4-8
Runs in
ap-northeast-1, ap-northeast-3

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
au.anthropic.claude-opus-4-8
Runs in
ap-southeast-2, ap-southeast-4

Inference stays in the single region the request was made to.

Model id
anthropic.claude-opus-4-8
Runs in
ap-northeast-1, ap-southeast-4, eu-north-1, eu-west-1, us-east-1

Google picks the region.

Model id
claude-opus-4-8
Runs in
Any region the vendor runs

Inference stays in the United States.

Model id
claude-opus-4-8
Runs in
United States

Inference stays in the European Union.

Model id
claude-opus-4-8
Runs in
European Union

Inference may run in any AWS region. Cheapest, with no residency guarantee.

Model id
global.openai.gpt-6-sol
Runs in
ap-east-2, ap-northeast-1, ap-northeast-2, ap-northeast-3, ap-south-1, ap-southeast-1, ap-southeast-2, ap-southeast-5, ap-southeast-7, ca-central-1, eu-central-1, eu-north-1, eu-west-1, eu-west-2, eu-west-3, sa-east-1, us-east-1, us-east-2, us-west-1, us-west-2

Inference stays inside one geography (US, EU, APAC, Japan or Australia).

Model id
us.openai.gpt-6-sol
Runs in
ca-central-1, us-east-1, us-east-2, us-west-1, us-west-2

The vendor's ordinary endpoint for this model.

Model id
gpt-6-sol
Runs in
Any region the vendor runs

Inference stays in the European Union.

Model id
gpt-6-sol
Runs in
European Union

Inference stays in the United States.

Model id
gpt-6-sol
Runs in
United States

The vendor's ordinary endpoint for this model.

Model id
gemini-2.5-pro
Runs in
Any region the vendor runs

Google picks the region.

Model id
gemini-2.5-pro
Runs in
Any region the vendor runs

Inference stays in the region the router names.

Model id
gemini-2.5-pro
Runs in
asia-northeast1, europe-central2, europe-north1, europe-southwest1, europe-west1, europe-west4, europe-west8, europe-west9, northamerica-northeast1, us-central1, us-east1, us-east4, us-east5, us-south1, us-west1, us-west4
May run here Vendor region, not used
A request runs only where the selected route allows. It never moves to another variant or region on its own.

Switch models instantly

Applications call a router, not a model. Change the router's model, or move an alias to another router, and new requests use it with no code change or deploy.

Your application
model="multi-router/support"
Never changes
Response headers
x-mr-upstream-model: anthropic/claude-sonnet-4.6x-mr-provider: anthropicx-mr-region: global
x-mr-upstream-model: anthropic/claude-sonnet-4.6x-mr-provider: bedrockx-mr-region: eu-central-1
x-mr-upstream-model: anthropic/claude-opus-4.8x-mr-provider: vertexx-mr-region: eu
x-mr-upstream-model: openai/gpt-6-solx-mr-provider: openaix-mr-region: us
x-mr-upstream-model: openai/gpt-6-solx-mr-provider: bedrockx-mr-region: us-east-1
x-mr-upstream-model: google/gemini-2.5-prox-mr-provider: geminix-mr-region: global
supportrouter

Select a model. The application keeps calling the same router.

Your keys, your rates, no lock-in

Routers call providers with your keys, so your contracts, rates and quotas apply. Each endpoint accepts its provider's request shape, so leaving is a base URL change.

One request, translatedreq_2Wd7PmRx4e
1 Your app sendsOpenAI Chat CompletionsPOST /v1/chat/completions · model "multi-router/support"
2 Multi-Router callsAWS Bedrock, in its native formateu.anthropic.claude-sonnet-4-6 · eu-central-1
3 Your app receivesAn OpenAI chat.completionStreaming, tool calls and usage in the shape it asked for
Any request shape can reach

Any request shape, any vendor

Call any model from the SDK you already use. Multi-Router translates the request into the vendor's format and the response back into yours.

LLM Provider KeysAdd key
Anthropic productionsk-ant-…a41c · global default
OpenAI EUsk-…9d2e · eu default
Bedrock EUAWS access key · eu-central-1 enabled
Vertex EUService account · europe-west4 enabled
Keys are encrypted at rest. Requests are billed to your provider accounts.

Your provider keys

Routers call each provider with your account, so your contracts, quotas and reserved capacity apply.

api.multi-router.airequest shapes
POST/v1/chat/completionsOpenAI
POST/v1/messagesAnthropic
POST/v1/responsesOpenAI Responses
POST/bedrock/model/{id}/invokeBedrock
POST/vertex/…:rawPredictVertex

Every provider's request shape

OpenAI, Anthropic, Bedrock and Vertex SDKs work unchanged.

Claude Sonnet 4.6bedrock · geo · eu-central-1 Negotiated
Per 1M tokensCatalogYour rate
Input $3.30 $2.64
Cache read $0.33 $0.264
Output $16.50 $13.20
One Claude Code turn $0.1547 $0.1238 -20%

Your negotiated rates

Set your own prices on any catalog offering. Costs and budgets use them.

acme/support-ft-8bCustom model
Served byAcme inference llm.acme.example/v1
Formatopenai_chat
Price per 1M$0.20 in · $0.60 out

Your own models

Self-hosted and fine-tuned models are routed and priced like catalog models.

A record of every routing decision

Each request records the key that sent it, the router, the vendor and region that served it, every attempt, its tokens, cost and latency.

multi-router.ai Routers Prompts Traffic
Models RS
Last 24 hours Router any Model any Vendor any Status any Request ID, correlation ID, customer or session Customize
Requests
1.28M4.2%
Success %
99.62%0.04%
5xx
31218%
Tokens
4.81B
Latency p95
2.41 s3.1%
Latency p99
5.87 s
Cost
$4,222.486.8%
Customers
1,846
Requestspeak 34,973 · 30m
14:3002:3014:22
5xxpeak 12 · 30m
14:3002:3014:22
Latency p95peak 3,370 ms · 30m
14:3002:3014:22
Costpeak $125.25 · 30m
14:3002:3014:22
TimeRequest IDRouterModelVendorRegionCustomerIn/OutCostLatencyStatus
14:22:07.412 req_6Tn1VbQe2s rt_support claude-sonnet-4.6 bedrock eu-central-1 cus_northwind 9,800 / 420 $0.019058 1,352 ms 200
req_6Tn1VbQe2s 200 claude-sonnet-4.6 bedrock View payload
Latency1,352ms
Cost$0.019057
Tokens9,800 → 420
TTFT412ms
Attempts2
Router overhead412µs

Request

Request ID
req_6Tn1VbQe2s
Timestamp
2026-10-20 14:22:07.412
Organization
org_YGhx6cFg1p
Token
key_7hQx · support-prod
Correlation ID
hd-7f3a91
Customer
cus_northwind
Session
ses_tk_48213
Peer IP
203.0.113.24

Routing

Alias
support
Router ID
rt_support
Model requested
multi-router/support
Model served
anthropic/claude-sonnet-4.6
Vendor
bedrock
Variant
geo
Region
eu-central-1
Wire id
eu.anthropic.claude-sonnet-4-6

Prompt binding

Prompt ID
pf_2Kq8Wv
Repository
acme/prompts
File path
prompts/support.md
Revision
4e1c9a7b

Latency

Total
1,352ms
TTFT
412ms
Request upload
3.0ms
Router overhead
412µs
Upstream
1,346ms
Router response
2.6ms

Tokens

Prompt
9,8007,000 cached700 cache-write700 at 5m
Completion
420
Request bytes
41,288
Response bytes
2,914

Cost

Total
$0.019057
Input
$0.006930
Output
$0.006930
Cache read
$0.002310
Cache write
$0.002888$0.002888 at 5m

Provider response

Request upload 3ms
Router · overhead 400µs
Upstream 1,346ms
Router · response 2.6ms
14:22:07.198 req_6Tn1VbQd8w rt_search gemini-2.5-flash vertex us-central1 cus_globex 4,200 / 350 $0.002196 611 ms 200
14:22:06.951 req_6Tn1VbQc3k rt_claude_code claude-opus-4.8 anthropic global priya.n@acme.example 64,120 / 2,310 $0.118240 8,904 ms 200
14:22:06.730 req_6Tn1VbQb1r rt_docs gpt-6-sol openai us cus_initech 63,120 / 1,880 $0.099131 4,215 ms 200
14:22:06.502 req_6Tn1VbQa7m rt_replies claude-sonnet-4.6 bedrock us-east-1 cus_umbrella — — 30,000 ms 504

High performance

Routing adds negligible latency because it is the fastest router on the planet: 47 ms at p99.

Router Request OverheadTraffic · Latency · last 24 hours
p509 ms
p9024 ms
p9531 ms
p9947 ms
45 ms30 ms15 ms
p99p95p50
One request req_6Tn1VbQe2s1,363 ms
Request upload 3.0 ms
Routing 11 ms
Upstream 1,346 ms
Router response 2.6 ms
Routing is measured on every request and reported in Traffic, separate from the provider's time.

Create a router

Create a router, set your SDK base URL to Multi-Router and send a request.

Start for Free