Models

Explore the hosted models available for direct Responses API calls and inference stacks. Each model supports a different context length and set of input modalities.

The prices and limits below describe Adaption’s hosted deployments. The linked upstream model cards may advertise different capabilities or limits.

Pricing

Prices are in credits per 1 million tokens. Cached input prices are shown in parentheses. The cached rate applies only to cached input tokens; all other input tokens use the standard rate.

Model ID	Input credits (cached)	Output credits
gemma-4-31b-it	0.09 (0.05)	0.34
glm-5.2	0.104 (0.10)	4.40
qwen3.5-397b	0.55 (0.225)	3.50
gemma-4-26b-a4b-it	0.0765 (0.0425)	0.255
qwen3.8-27b	0.039 (0.0039)	1.25
deepseek-v4-flash	0.04444 (0.001414)	0.13332
glm-53-flash	0.15 (0.03)	0.50

Configurations

Context length is the maximum combined number of input and output tokens supported by a model.

Model ID	Context length (tokens)	Input modalities
gemma-4-31b-it	256,000	Text, image
glm-5.2	786,432	Text
qwen3.5-397b	131,072	Text, image
gemma-4-26b-a4b-it	256,000	Text, image
qwen3.8-27b	262,144	Text, image
deepseek-v4-flash	1,048,576	Text
glm-53-flash	1,048,576	Text

To send images to a model that supports them, see Send image inputs.

Call a model or stack

Pass a model ID as model to call a hosted model directly. To call an inference stack, use stack:<slug>. The stack supplies its configured model, tools, datasets, and memory setting.

from adaption import Adaption

client = Adaption()  # reads ADAPTION_API_KEY

response = client.responses.create(
    model="stack:your-stack-slug",
    input="Hello",
)

print(response.output_text)

Use a stack’s slug when possible. stack:<uuid> remains supported for existing callers.

List available models

Model availability can vary by API key. Pass type="hosted" to list the hosted models available to your key without including inference stacks or harnesses:

from adaption import Adaption

client = Adaption()

for model in client.models.list(type="hosted").data:
    print(model.id)