LLM Routermodel = "auto"
Not every request needs the biggest, priciest model
Smart routing selects the best model for every request in real time — cutting costs significantly without compromising task quality. Cache-friendly by design: each session automatically sticks to the same model, so prompt caches are never lost.
No extra setup required: session identifiers from popular agents like Claude Code, Codex, and OpenCode are recognized automatically. For precise control over session scope, add this to your request:X-Aihubmix-Session-Id: your-session-id
Scores span 5 major categories and 23 sub-dimensions, normalized from a blend of internal evaluations and public leaderboards.
API referencechat.py
import openai client = openai.OpenAI( api_key="<AIHUBMIX_API_KEY>", base_url="https://aihubmix.com/v1" ) response = client.chat.completions.create( model="auto", # bare auto = auto:cost_optimized (default) messages=[{"role": "user", "content": "Hello"}] ) print(response.choices[0].message.content)
Routing strategy
Routing strategies
Model coverage across the four routing strategies
Text › Overall
Weighted by "Cost first"Model
18
claude-opus-4-6-thinkAnthropic
86.7
$2.5 / $51,673 ms21
claude-opus-4-7-thinkAnthropic
86.3
$2.5 / $53,036 ms14
claude-opus-4-6Anthropic
85.4
$2.5 / $5822 ms19
claude-opus-4-7Anthropic
84.0
$2.5 / $5719 ms26
gemini-3.1-pro-previewGoogle
81.7
$1 / $66,816 ms20
gemini-3.5-flashGoogle
79.9
$0.75 / $64,328 ms28
gpt-5.4-highOpenAI
79.4
$1.25 / $6—24
claude-opus-4-8Anthropic
79.0
$2.5 / $51,591 ms29
gpt-5.5OpenAI
78.5
$2.5 / $64,766 ms8
grok-4.20-beta-0309-reasoningxAI
78.5
$1 / $31,685 ms11
claude-sonnet-4-6Anthropic
77.8
$1.5 / $5995 ms12
claude-sonnet-4-6-thinkAnthropic
77.3
$1.5 / $51,762 ms9
glm-5.2Z.ai
76.8
$0.563 / $3.52,770 ms1
xiaomi-mimo-v2.5-proCurrent pickXiaomi
76.1
$0.24 / $21,673 ms10
qwen3.7-plusQwen
75.0
$0.141 / $44,873 ms13
qwen3.6-max-previewQwen
74.4
$0.634 / $6651 ms4
grok-4-1-fast-non-reasoningxAI
74.2
$0.1 / $2.51,188 ms2
deepseek-v4-proDeepSeek
73.5
$0.232 / $22,037 ms17
gpt-5.4-miniOpenAI
71.0
$0.375 / $61,354 ms6
grok-4.3xAI
69.7
$0.625 / $21,224 ms22
qwen3.6-plusQwen
69.5
$0.141 / $64,120 ms16
qwen3.5-397b-a17bQwen
69.4
$0.082 / $62,106 ms3
deepseek-v4-flashDeepSeek
67.3
$0.077 / $2879 ms5
xiaomi-mimo-v2.5Xiaomi
66.6
$0.078 / $21,673 ms15
gemini-3.1-flash-liteGoogle
66.4
$0.125 / $6733 ms7
grok-4-1-fast-reasoningxAI
66.0
$0.1 / $2.52,003 ms30
qwen3.5-122b-a10bQwen
62.0
$0.056 / $8961 ms31
qwen3.5-27bQwen
59.7
$0.042 / $82,085 ms25
gpt-5.4-nanoOpenAI
57.8
$0.1 / $6.25702 ms33
qwen3.5-flashQwen
56.5
$0.014 / $105,775 ms32
qwen3.5-35b-a3bQwen
55.9
$0.028 / $83,310 ms23
gpt-oss-120bOpenAI
43.6
$0.09 / $5366 ms27
gpt-oss-20bOpenAI
33.5
$0.055 / $5120 msSources: internal evals and public leaderboards, jointly normalized.
