Gallery
A router you call like a model. Send one model id and each request goes to the right model.
Picking a model should be a per-request decision that almost nobody makes. Currently, it gets made once at integration time and then forgotten about until you run out of credit or look at your AI spend. The result is a frontier model answering questions like "what's 5 km in miles", and a small model failing the question that needed the frontier one.
Gallery is our router. Set model to telluvian/gallery-1 and each request
goes to the cheapest model that still clears a quality bar you set. The response
gives you the answer and reports which model actually ran.
{
"model": "telluvian/gallery-1",
"messages": [{"role": "user", "content": "Reformat this address as JSON: ..."}],
"include_scores": false
}Everything else about the request is unchanged. Streaming, multi-turn messages and the Responses API work exactly as they do when you name a model yourself.
Gallery runs on Model Select
The choice itself is made by Model Select, which reads your prompt and predicts how well each model in the zoo will handle it. Gallery asks Model Select for a model, then sends your request there. If you would rather make the model call yourself, call Model Select directly.
Verification is opt-in
Gallery does not turn on verification. Set include_scores: true to add
verification.
Sending a request
curl https://api.telluvian.ai/v1/chat/completions \
-H "Authorization: Bearer $TELLUVIAN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "telluvian/gallery-1",
"messages": [{"role": "user", "content": "Summarise this changelog in one line: ..."}],
"xPerf": "openai/gpt-6.1-terra",
"include_scores": false
}'Reading what ran
telluvian/gallery-1 is not a model. It is resolved to a real one before your
request is sent anywhere, so the model field on the response, and on every
chunk of a stream, names the model that generated the text:
{
"id": "chatcmpl-...",
"model": "google/gemini-3.8-flash",
"sessionId": "3f2b7c58-9d41-4e0a-9a7c-6f0b1c2d3e4f",
"choices": [ ... ]
}Log response.model to track what generated each answer. Non-streamed
responses also carry the sessionId Model Select used. See
keeping a conversation together.
The choice is made before the model is called
Nothing is decided mid-generation, so stream: true behaves normally: the
same SSE chunks in the same order. The one difference you can measure is time
to first byte, because choosing costs a round trip before your request leaves
for the model.
Steering the choice
Gallery accepts the Model Select controls on
the request body alongside model: xPerf, xPerfEffort, xLatency,
modelZoo and sessionId. Gallery reads the controls only when model is
telluvian/gallery-1, passes them to Model Select unchanged, and never forwards
them to the model.
Controls in the model id
Some tools only let you set a model name: agent frameworks, chat UIs, and config files that take a model string and nothing else. For those, put the controls in the model id after a colon:
telluvian/gallery-1:xPerf=0.9
telluvian/gallery-1:xPerf=0.9,xLatency=instant
telluvian/gallery-1:xPerf=anthropic/claude-sonnet-5-5,xPerfEffort=highThis is the same request as sending the controls on the body. Gallery routes
it, bills it and reports it the same way, and the response's model still
names the model that ran. It works on Chat Completions and the
Responses API, and OpenAI-compatible clients pass the model
id through unchanged:
# No extra_body needed: the controls travel in the model id.
response = client.chat.completions.create(
model="telluvian/gallery-1:xPerf=0.9",
messages=[{"role": "user", "content": "Summarise this changelog in one line: ..."}],
)
print(response.model) # the model that actually ranThe rules:
xPerf,xPerfEffortandxLatencycan go in the model id. Separate several with commas.modelZooandsessionIdstay on the body:modelZoouses commas itself, and asessionIdchanges with each conversation, which a configured model name cannot.- Names are case-insensitive.
XPerf=0.9works, and so doxLatency=InstantandxPerfEffort=HIGH. - A number in the model id is a number.
xPerf=0.9sets the quality bar to0.9, so the quoting problem a JSON body can have does not arise. Anything else, such asxPerf=anthropic/claude-sonnet-5-5, is read as a model id. - Set each control in one place. A control in both the model id and the
body must have the same value in each. If the two differ, the request is a
400. - Mistakes are a
400before anything runs. A misspelt or unsupported control, a control given twice, or a suffix that is notkey=valueis rejected with a message showing the expected format, and nothing is billed. A suffix on any model other thantelluvian/gallery-1is an unknown model, so a404.
GET /v1/models/{id} accepts these ids too, for tools that check a model
exists before using it, and returns the Gallery entry under the id you asked
for. GET /v1/models lists only the plain telluvian/gallery-1.
What it costs
A Gallery request costs the selected model's list price plus the Model Select fee. Naming a model yourself skips Model Select and the fee. See Pricing.
Your invoice names real models, never this id
telluvian/gallery-1 has no token rate of its own and never appears as a line
item. Model spend is recorded against the model that generated the tokens, so
usage reporting keeps working without knowing that Gallery is in play. The
Model Select fee is billed as part of that model's charge.
When Gallery is unavailable
If Model Select cannot make the choice, the request fails with a 503 and
nothing is charged. It is never served by quietly substituting a model. A
Gallery request that silently became something else would be indistinguishable
from a working one, and you would find out from the bill.
One narrow substitution does happen. If the model chosen is not one the catalogue can serve, the default model runs instead. The response names it like any other.