Tasco Inference

Inference for open-weight models

An OpenAI-compatible API that serves open-weight models from their official weights, streams tokens as they are generated, and keeps no copy of your prompts or outputs.

Models

Model	Weights	Context	Input	Cached input	Output
Qwen3.6 35B A3B	official FP8 release	65,536	$0.10	$0.05	$0.95

Prices in USD per million tokens. The live list, with limits and supported parameters, is at api.tascoinference.com/v1/models.

How we serve

• Official weights published by the model authors, served unmodified with vLLM: no pruning, no further quantization.
• Streaming responses with exact token usage, including cached prompt tokens.
• When capacity is full we answer 429 immediately instead of queueing, so requests never wait silently.
• Tool calling, structured outputs and reasoning are supported.

Your data

Prompts and outputs are processed in memory only for the time needed to answer the request. We do not log or store them and we never use them to train models. Inference runs on GPUs rented from third-party cloud providers. Details are in the privacy policy.

Contact

support@tascoinference.com