Let the gateway pick
Send minirouter/auto and the gateway picks the model. The x-minirouter-model header names the one that answered.
curl https://api.minirouter.sh/v1/chat/completions \
-H "Authorization: Bearer $MINIROUTER_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"minirouter/auto","messages":[{"role":"user","content":"ping"}]}'
# x-minirouter-model: <the model that answered>How the pick is made
- 1
IntentAdd a suffix to choose a list yourself:
:cheap,:code,:roleplay. Only the request shape is read, never the prompt.Bare Auto is Smart. Tools or structured output select Code.
- 2
Walk the listEach list is hand-ordered and changes only by a reviewed code change. Not combinable with your own providerOptions.gateway.models fallback list.
Best first, in the order shown below.
- 3
Skip what failsTools, structured output and long prompts can push a candidate over the price bound for your exact request.
Catalog, allowlist, cost tier, route health, price bound.
- 4
AnswerOne worst-case hold covers the eligible set. A failure before output tries the next candidate on Chat Completions and Responses.
The first pass answers.
x-minirouter-modelnames it.
Nothing passes: 503 auto_unresolvable, uncharged.
Current candidates
Platform availability before your account rules and request shape apply.
Raw table: /api/v1/auto · refreshed every minute
AutoDefault
minirouter/autoDocs →Picks a strong model for the request from its structure: coding agents get the code list, a low cost tier starts from the cheap list, everything else takes smart.Prices include the total 8% routing markup over provider cost.
Default
Tools
Low cost tier
Auto Cheap
minirouter/auto:cheapDocs →The least expensive model that still answers well.Prices include the total 8% routing markup over provider cost.
| # | Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|---|
| 1 | xiaomi/mimo-v2.5 | $0.151 | $0.302 | 1.1M | vercel | First available |
| 2 | xiaomi/mimo-v2.5-pro | $0.47 | $0.94 | 1.1M | vercel | Next in line |
| 3 | tencent/hy3 | $0.151 | $0.626 | 262K | vercel | Next in line |
| 4 | deepseek/deepseek-v4-flash-0731 | $0.082 | $0.165 | 1M | vercel | Next in line |
| 5 | zai/glm-5.3-flash | $0.162 | $0.54 | 1M | vercel | Next in line |
Auto Code
minirouter/auto:codeDocs →A model with dependable tool calling and long context for coding agents.Prices include the total 8% routing markup over provider cost.
| # | Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|---|
| 1 | zai/glm-5.3-flash | $0.162 | $0.54 | 1M | vercel | First available |
| 2 | openai/gpt-6-sol | $2.16 | $10.80 | 1.1M | vercel | Next in line |
| 3 | spacexai/grok-4.6 | $2.16 | $6.48 | 500K | vercel | Next in line |
| 4 | google/gemini-3.8-flash | $0.81 | $4.05 | 1M | vercel | Next in line |
| 5 | openai/gpt-6-luna | $0.108 | $0.54 | 1.1M | vercel | Next in line |
Auto Roleplay
minirouter/auto:roleplayDocs →A high-intelligence, long-context model with low input cost for repeatedly sent conversations.Prices include the total 8% routing markup over provider cost.
| # | Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|---|
| 1 | zai/glm-5.3-flash | $0.162 | $0.54 | 1M | vercel | First available |
| 2 | openai/gpt-6-luna | $0.108 | $0.54 | 1.1M | vercel | Next in line |
| 3 | deepseek/deepseek-v4-flash-0731 | $0.082 | $0.165 | 1M | vercel | Next in line |
| 4 | google/gemini-3.8-flash | $0.81 | $4.05 | 1M | vercel | Next in line |
Auto Smart
minirouter/auto:smartDocs →A strong general-purpose model at a mid-range price.Prices include the total 8% routing markup over provider cost.
| # | Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|---|
| 1 | zai/glm-5.3-flash | $0.162 | $0.54 | 1M | vercel | First available |
| 2 | openai/gpt-6-sol | $2.16 | $10.80 | 1.1M | vercel | Next in line |
| 3 | spacexai/grok-4.6 | $2.16 | $6.48 | 500K | vercel | Next in line |
| 4 | google/gemini-3.8-flash | $0.81 | $4.05 | 1M | vercel | Next in line |
| 5 | openai/gpt-6-luna | $0.108 | $0.54 | 1.1M | vercel | Next in line |
FusionA classifier labels the request. The label picks the model and reasoning effort. Only this ID reads your request.
Reads the request: a classifier (TypeSafe Jev) labels its task type and complexity, and that label picks the model and reasoning effort.Prices include the total 10% routing markup over provider cost.
minirouter/fusion
- Code · Simple
- zai/glm-5.3-flashdeepseek/deepseek-v4-pro-0813maxgoogle/gemini-3.8-flashhigh
- Code · Everyday
- google/gemini-3.8-flashhighzai/glm-5.3-flashspacexai/grok-4.6high
- Code · Complex
- google/gemini-3.8-flashhighspacexai/grok-4.6highanthropic/claude-fable-5.1
- Code · Expert
- anthropic/claude-fable-5.1spacexai/grok-4.6highopenai/gpt-6-astramedium
- Reasoning · Simple
- xiaomi/mimo-v2.6-prozai/glm-5.3-flashstepfun/step-5-preview
- Reasoning · Everyday
- anthropic/claude-opus-5.5xhighxiaomi/mimo-v2.6-prozai/glm-5.3-flash
- Reasoning · Complex
- anthropic/claude-opus-5.5xhighopenai/gpt-6-astramediumxiaomi/mimo-v2.6-pro
- Reasoning · Expert
- anthropic/claude-opus-5.5xhighxiaomi/mimo-v2.6-prozai/glm-5.3-flash
- Writing · Simple
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astralowdeepseek/deepseek-v4.1-flashmax
- Writing · Everyday
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumdeepseek/deepseek-v4.1-flashmax
- Writing · Complex
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumdeepseek/deepseek-v4.1-flashmax
- Writing · Expert
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumgoogle/gemini-3.8-flashhigh
- Conversation · Simple
- openai/gpt-6-astralowdeepseek/deepseek-v4.1-flashmaxanthropic/claude-opus-5.5low
- Conversation · Everyday
- openai/gpt-6-astramediumdeepseek/deepseek-v4.1-flashmaxanthropic/claude-opus-5.5low
- Conversation · Complex
- openai/gpt-6-astramediumdeepseek/deepseek-v4.1-flashmaxanthropic/claude-opus-5.5low
- Conversation · Expert
- openai/gpt-6-astramediumanthropic/claude-opus-5.5lowgoogle/gemini-3.8-flashhigh
- No labelUsed when the classifier cannot answer.
- openai/gpt-6-astramediumdeepseek/deepseek-v4.1-flashmaxanthropic/claude-opus-5.5low
| Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|
| zai/glm-5.3-flash | $0.165 | $0.55 | 1M | vercel | Available |
| deepseek/deepseek-v4-pro-0813 | $0.726 | $2.18 | 1M | vercel | Available |
| google/gemini-3.8-flash | $0.825 | $4.13 | 1M | vercel | Available |
| spacexai/grok-4.6 | $2.20 | $6.60 | 500K | vercel | Available |
| anthropic/claude-fable-5.1 | $11.00 | $55.00 | 1M | vercel | Available |
| openai/gpt-6-astra | $11.00 | $55.00 | 1.1M | vercel | Available |
| xiaomi/mimo-v2.6-pro | $0.478 | $0.957 | 1M | vercel | Available |
| stepfun/step-5-preview | $1.10 | $2.97 | 1M | vercel | Available |
| anthropic/claude-opus-5.5 | $4.40 | $22.00 | 1M | vercel | Available |
| deepseek/deepseek-v4.1-flash | $0.33 | $1.32 | 1M | vercel | Available |
Fusion CheapA classifier labels the request. The label picks the model and reasoning effort. Only this ID reads your request.
Fusion Cheap: the classifier selects a model and effort from the published price-quality frontier.Prices include the total 10% routing markup over provider cost.
minirouter/fusion:cheap
- Code · Everyday
- zai/glm-5.3-flashtencent/hy3minimax/minimax-m3
- Code · Complex
- zai/glm-5.3-flashdeepseek/deepseek-v4-pro-0813maxgoogle/gemini-3.8-flashhigh
- Code · Expert
- zai/glm-5.3-flashgoogle/gemini-3.8-flashhighspacexai/grok-4.6high
- Reasoning · Simple
- inclusionai/ling-3.0-flashxiaomi/mimo-v2.6-flashdeepseek/deepseek-v4.1-flashnone
- Reasoning · Everyday
- xiaomi/mimo-v2.6-flashzai/glm-5.3-flashdeepseek/deepseek-v4.1-flashmax
- Reasoning · Complex
- zai/glm-5.3-flashxiaomi/mimo-v2.6-prodeepseek/deepseek-v4.1-flashmax
- Reasoning · Expert
- xiaomi/mimo-v2.6-prometa/muse-spark-1.3xhighanthropic/claude-opus-5.5medium
- Writing · Simple
- inclusionai/ling-3.0-flashdeepseek/deepseek-v4.1-flashnonegoogle/gemini-3.8-flashhigh
- Writing · Everyday
- deepseek/deepseek-v4.1-flashmaxgoogle/gemini-3.8-flashhighanthropic/claude-opus-5.5low
- Writing · Complex
- deepseek/deepseek-v4.1-flashmaxgoogle/gemini-3.8-flashhighanthropic/claude-opus-5.5low
- Writing · Expert
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astralowinclusionai/ling-3.0-flash
- Conversation · Simple
- inclusionai/ling-3.0-flashdeepseek/deepseek-v4.1-flashnonespacexai/grok-4.6low
- Conversation · Everyday
- deepseek/deepseek-v4.1-flashmaxspacexai/grok-4.6lowopenai/gpt-5.6-terramedium
- Conversation · Complex
- deepseek/deepseek-v4.1-flashmaxgoogle/gemini-3.8-flashhighanthropic/claude-opus-5.5low
- Conversation · Expert
- openai/gpt-6-astralowinclusionai/ling-3.0-flashamazon/nova-micro
- No labelUsed when the classifier cannot answer.
- deepseek/deepseek-v4.1-flashmaxspacexai/grok-4.6lowopenai/gpt-5.6-terramedium
| Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|
| inclusionai/ling-3.0-flash | $0.023 | $0.069 | 256K | vercel | Available |
| zai/glm-5.3-flash | $0.165 | $0.55 | 1M | vercel | Available |
| tencent/hy3 | $0.154 | $0.638 | 262K | vercel | Available |
| minimax/minimax-m3 | $0.33 | $1.32 | 512K | vercel | Available |
| deepseek/deepseek-v4-pro-0813 | $0.726 | $2.18 | 1M | vercel | Available |
| google/gemini-3.8-flash | $0.825 | $4.13 | 1M | vercel | Available |
| spacexai/grok-4.6 | $2.20 | $6.60 | 500K | vercel | Available |
| xiaomi/mimo-v2.6-flash | $0.154 | $0.308 | 1M | vercel | Available |
| deepseek/deepseek-v4.1-flash | $0.33 | $1.32 | 1M | vercel | Available |
| xiaomi/mimo-v2.6-pro | $0.478 | $0.957 | 1M | vercel | Available |
| meta/muse-spark-1.3 | $1.38 | $4.68 | 1M | vercel | Available |
| anthropic/claude-opus-5.5 | $4.40 | $22.00 | 1M | vercel | Available |
| openai/gpt-6-astra | $11.00 | $55.00 | 1.1M | vercel | Available |
| openai/gpt-5.6-terra | $2.20 | $13.20 | 1.1M | vercel | Available |
| amazon/nova-micro | $0.039 | $0.154 | 128K | vercel | Available |
Fusion MaxA classifier labels the request. The label picks the model and reasoning effort. Only this ID reads your request.
Fusion Max: the classifier selects a model and effort from the published price-quality frontier.Prices include the total 10% routing markup over provider cost.
minirouter/fusion:max
- Code · Simple
- anthropic/claude-fable-5.1spacexai/grok-4.6highopenai/gpt-6-astramedium
- Code · Everyday
- anthropic/claude-fable-5.1spacexai/grok-4.6highopenai/gpt-6-astramedium
- Code · Complex
- anthropic/claude-fable-5.1spacexai/grok-4.6highopenai/gpt-6-astramedium
- Code · Expert
- anthropic/claude-fable-5.1spacexai/grok-4.6highopenai/gpt-6-astramedium
- Reasoning · Simple
- anthropic/claude-opus-5.5xhighopenai/gpt-6-astramediumxiaomi/mimo-v2.6-pro
- Reasoning · Everyday
- anthropic/claude-opus-5.5xhighopenai/gpt-6-astramediumxiaomi/mimo-v2.6-pro
- Reasoning · Complex
- anthropic/claude-opus-5.5xhighopenai/gpt-6-astramediumxiaomi/mimo-v2.6-pro
- Reasoning · Expert
- anthropic/claude-opus-5.5xhighopenai/gpt-6-astraxhighxiaomi/mimo-v2.6-pro
- Writing · Simple
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumgoogle/gemini-3.8-flashhigh
- Writing · Everyday
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumgoogle/gemini-3.8-flashhigh
- Writing · Complex
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumgoogle/gemini-3.8-flashhigh
- Writing · Expert
- anthropic/claude-opus-5.5mediumopenai/gpt-6-astramediumgoogle/gemini-3.8-flashhigh
- Conversation · Simple
- openai/gpt-6-astramediumanthropic/claude-opus-5.5lowgoogle/gemini-3.8-flashhigh
- Conversation · Everyday
- openai/gpt-6-astramediumanthropic/claude-opus-5.5lowgoogle/gemini-3.8-flashhigh
- Conversation · Complex
- openai/gpt-6-astramediumanthropic/claude-opus-5.5lowgoogle/gemini-3.8-flashhigh
- Conversation · Expert
- openai/gpt-6-astramediumanthropic/claude-opus-5.5lowgoogle/gemini-3.8-flashhigh
- No labelUsed when the classifier cannot answer.
- openai/gpt-6-astramediumanthropic/claude-opus-5.5lowgoogle/gemini-3.8-flashhigh
| Model | Input $/M | Output $/M | Context | Providers | Status |
|---|---|---|---|---|---|
| anthropic/claude-fable-5.1 | $11.00 | $55.00 | 1M | vercel | Available |
| spacexai/grok-4.6 | $2.20 | $6.60 | 500K | vercel | Available |
| openai/gpt-6-astra | $11.00 | $55.00 | 1.1M | vercel | Available |
| anthropic/claude-opus-5.5 | $4.40 | $22.00 | 1M | vercel | Available |
| xiaomi/mimo-v2.6-pro | $0.478 | $0.957 | 1M | vercel | Available |
| google/gemini-3.8-flash | $0.825 | $4.13 | 1M | vercel | Available |
Your account rules
Set once on Routing. Every key on the account follows them.
Cost tierCaps the output price per million tokens of any candidate Auto may pick. Not a request budget. An explicit intent keeps its list under the same cap.
Low
$1
per M output tokens
Bare Auto starts from Cheap.
Medium
$5
per M output tokens
Bare Auto starts from Smart.
High
No cap
the default
Bare Auto starts from Smart.
Allowed modelsApplies to requests that name a model too. Adding a model here does not add it to Auto's lists.
Exact IDs or
openai/*. Keys can narrow it, requests cannot widen it.Prevent overridesResponds with
x-minirouter-overrides: ignored. Allowed models and the cost tier apply either way.Ignores a request’s
providerOptions.gateway.Provider sortA sort set on the request wins unless Prevent overrides is on.
Balanced, price, latency or throughput. Orders providers, not models.
Default modelAuto, an intent, or any catalog model. A model named on the request is left as sent. Dashboard chat uses its own picker.
What integration keys get when
modelis omitted.