Console

Qwen3.6 27B

Context
262K
Max output
262K
Input → output
text, image → text
Weights
Open
Providers
4
01

Providers

4 providers · list price, latency and uptime
ProviderInput /1MCached /1MOutput /1MTTFB p50Uptime 24hRetention
1
DeepInfra
$0.32–$3.20– – none
Context
262K
Max output
262K
Precision
fp8
Regions
default
Headquarters
–
Data retention
Zero retention
2
StreamLake
$0.60–$3.60– – may retain
Context
262K
Max output
66K
Precision
–
Regions
default
Headquarters
–
Data retention
May retain prompts
Supported parameters
response_format
3
Chutes
$0.25$0.025$2.00– – none
Context
262K
Max output
66K
Precision
fp8
Regions
default
Headquarters
–
Data retention
Zero retention
4
Phala
$0.32$0.03$3.25– – none
Context
178K
Max output
178K
Precision
–
Regions
default
Headquarters
–
Data retention
Zero retention
Supported parameters
response_format

By default the router tries our listings in routing order and fails over before the first byte — never mid-answer. Pin, exclude or order providers with a routing policy. Each provider charges its own price: name one with provider.only and you pay its price; otherwise you pay the price of the provider that serves the request ($0.32 in / $3.20 out per 1M where routing starts, unless it fails over). provider.sort: "price" tries the provider with the lowest input + output price first, and provider.max_price leaves out a provider above your limit. Uptime is our own 24h measurement; a dash means too few requests to measure yet. Open a row for its specs, parameters and policies.

Phala: Phala serves this model at its own price, as any other provider does: a request that names no provider can be sent to it, and a request that names it with provider.only or provider.order chooses it.

Phala's capacity is smaller than other providers'. A long prompt, or one without max_tokens, can be refused with a 413 that waiting will not fix: shorten the prompt, set max_tokens, or allow another provider.

Phala runs the model in a hardware-isolated enclave (TEE) on one of several third-party TEE backends that Phala picks; RouterPlus neither chooses nor sees which one serves a call. Your prompt passes through the RouterPlus gateway before it reaches Phala, so the enclave protects it at Phala, not on our side. Zero data retention is requested from Phala on every call: it is Phala's commitment, not something RouterPlus measures. Phala publishes an attestation report and signs a receipt for each answer; RouterPlus does not check them per request.

02

Quickstart

any SDK — only the base URL changes
curl -N https://api.routerplus.com/v1/chat/completions \
  -H "Authorization: Bearer $TM_API_KEY" -H "content-type: application/json" \
  -d '{"model":"qwen/qwen3.6-27b","stream":true,"max_tokens":256,
       "messages":[{"role":"user","content":"hi"}]}'
03

Pricing

per 1M tokens
prompt$0.32
completion$3.2

Other providers charge from $0.25 / $2.00 in / out per 1M: name one with provider.only to pay its price.

Every billed response carries usage.cost — recompute your bill from the wire.