Skip to main content

ResetData

ResetData is an Australian sovereign AI-inference platform with OpenAI-compatible chat completions and audio transcription APIs. ResetData model ids are provider/model slugs, so full LiteLLM model ids have three segments, for example resetdata/zai/glm-5.3. LiteLLM removes only the resetdata/ prefix before it sends the model value to ResetData.

Since v1.112.0

Provider support was added in PR #50819. Use this version or a later build that includes the provider.

Overview​

PropertyDetails
Provider prefixresetdata/
API key environment variableRESETDATA_API_KEY
Default API basehttps://api.resetdata.ai/api/v1
Optional API base overrideRESETDATA_API_BASE or the api_base parameter
Upstream endpoints/chat/completions, /audio/transcriptions
Provider documentationResetData docs

Setup​

Create an API key in the ResetData platform and export it before you start the SDK or Gateway:

export RESETDATA_API_KEY="your-resetdata-api-key"

Pick a model from the ResetData catalog and prepend the resetdata/ prefix, for example resetdata/zai/glm-5.3.

LiteLLM Python SDK​

from litellm import completion

response = completion(
model="resetdata/zai/glm-5.3",
messages=[{"role": "user", "content": "Say hello in five words"}],
max_tokens=2000,
)

print(response.choices[0].message.content)

For streaming, pass stream=True and iterate over the response chunks:

from litellm import completion

response = completion(
model="resetdata/zai/glm-5.3",
messages=[{"role": "user", "content": "Say hello in five words"}],
max_tokens=2000,
stream=True,
)

for chunk in response:
print(chunk.choices[0].delta.content or "", end="")

Several ResetData models support reasoning, and reasoning tokens count toward max_tokens. Leave enough room for both, or the response can end before any visible text.

LiteLLM accepts reasoning_effort only for the reasoning models in the table below. For any other ResetData model, including one ResetData lists with reasoning that isn't in the table, LiteLLM rejects reasoning_effort with UnsupportedParamsError before it calls ResetData. Set drop_params=True on the call, or drop_params: true under litellm_settings on the Gateway, to drop the parameter and send the rest of the request. To send reasoning_effort to ResetData anyway, pass allowed_openai_params=["reasoning_effort"] on the call, or set allowed_openai_params: ["reasoning_effort"] in the deployment's litellm_params. ResetData answers 400 if the model doesn't support it.

LiteLLM Gateway​

Install the proxy with pip install 'litellm[proxy]'. Set RESETDATA_API_KEY on the Gateway process, and set LITELLM_MASTER_KEY to a strong key for Gateway clients. The ResetData key authenticates upstream requests. The master key authenticates your client to LiteLLM.

export RESETDATA_API_KEY="your-resetdata-api-key"
export LITELLM_MASTER_KEY="sk-your-strong-master-key"

Save this configuration as config.yaml:

model_list:
- model_name: resetdata-glm-5.3
litellm_params:
model: resetdata/zai/glm-5.3
api_key: os.environ/RESETDATA_API_KEY

general_settings:
master_key: os.environ/LITELLM_MASTER_KEY

Start the proxy:

litellm --config config.yaml --port 4000

In a second terminal with LITELLM_MASTER_KEY set, send a request using the configured model alias:

curl http://localhost:4000/v1/chat/completions \
-H "Authorization: Bearer $LITELLM_MASTER_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "resetdata-glm-5.3",
"messages": [{"role": "user", "content": "Say hello in five words"}],
"max_tokens": 2000
}'

The proxy returns an OpenAI-format chat completion. The usage block reports reasoning tokens separately:

{
"choices": [{"message": {"role": "assistant", "content": "Hello there, my dear friend!"}}],
"usage": {
"completion_tokens": 15,
"prompt_tokens": 17,
"total_tokens": 32,
"completion_tokens_details": {"reasoning_tokens": 7, "text_tokens": 8}
}
}

LiteLLM also accepts Anthropic Messages and Responses API requests for ResetData models through its translation layer. These requests use ResetData's chat completions endpoint, not native ResetData Messages or Responses endpoints.

Audio Transcription​

ResetData serves openai/whisper-large-v3-turbo on its OpenAI-compatible /audio/transcriptions endpoint. Add it to the Gateway configuration:

model_list:
- model_name: resetdata-whisper
litellm_params:
model: resetdata/openai/whisper-large-v3-turbo
api_key: os.environ/RESETDATA_API_KEY

Then send an audio file to the Gateway:

curl http://localhost:4000/v1/audio/transcriptions \
-H "Authorization: Bearer $LITELLM_MASTER_KEY" \
-F model=resetdata-whisper \
-F file=@speech.wav

The Gateway returns {"text": "..."} with the transcript. From the SDK, call litellm.transcription(model="resetdata/openai/whisper-large-v3-turbo", file=open("speech.wav", "rb")).

Supported Models​

These ResetData models are in the LiteLLM catalog with their context windows, output caps and prices, so LiteLLM tracks spend for them in USD. ResetData bills in Australian dollars, excluding GST. The catalog prices are ResetData's AUD prices converted at 1 AUD = 0.67566 USD, derived from the European Central Bank's euro reference rates for November 16, 2026. So tracked spend drifts from your ResetData invoice as the AUD moves, and it leaves out the GST your invoice adds.

ModelContext windowMax output tokensReasoningVision
resetdata/zai/glm-5.3524,28832,768YesNo
resetdata/qwen/qwen3.8-27b262,14432,768YesYes
resetdata/qwen/qwen3-coder-30b-a3b-instruct262,14465,536NoNo
resetdata/google/gemma-4-26b-a4b-it262,1448,192NoYes
resetdata/deepseek/deepseek-v4-flash1,048,57665,536YesNo

resetdata/openai/whisper-large-v3-turbo is in the catalog too. ResetData prices audio per minute of input, and LiteLLM tracks it per second of audio at the same converted rate.

resetdata/qwen/qwen3.8-27b takes images as base64 data URLs. ResetData answers an image URL on that model with a 500 error after about two minutes, while resetdata/google/gemma-4-26b-a4b-it takes both.

To bill at your own ResetData rates instead, set per-token costs in USD on the deployment. On a model with prompt caching, like resetdata/zai/glm-5.3 and resetdata/deepseek/deepseek-v4-flash, also set cache_read_input_token_cost, or LiteLLM bills cached tokens at the catalog rate. Give the deployment its own model_name, since LiteLLM load-balances deployments that share one. The values below are an example, not ResetData's prices:

model_list:
- model_name: resetdata-glm-5.3-own-rates
litellm_params:
model: resetdata/zai/glm-5.3
api_key: os.environ/RESETDATA_API_KEY
input_cost_per_token: 0.0000006
output_cost_per_token: 0.0000022
cache_read_input_token_cost: 0.0000001
LiteLLM Enterprise
SSO/SAML, audit logs, spend tracking, multi-team management, and guardrails, built for production.
Learn more →