Model library

ResetData's serverless catalogue offers leading open-weight models with high-speed inferencing, hosted entirely on sovereign Australian infrastructure.

Available models at a glance

Model Provider Type Architecture Context Licence Best for
Z.ai GLM 5.2zai/glm-5.2 Z.ai (Zhipu AI) Text, reasoning MoE: 753B total, 8 of 256 experts active per token 800K tokens MIT Complex reasoning, long-horizon agentic work, bilingual English/Mandarin
Gemma 4 26B A4B Instructgoogle/gemma-4-26b-a4b-it Google DeepMind Text, image and video input MoE: 26B total, ~4B active per token 256K tokens Gemma Terms of Use Long documents, RAG, document and image understanding, regulated data
Qwen3 Coder 30B A3B Instructqwen/qwen3-coder-30b-a3b-instruct Alibaba Text, code MoE: 30B total, ~3B active per token 256K tokens (1M with YaRN) Apache 2.0 Agentic coding at repository scale
Whisper Large V3 Turboopenai/whisper-large-v3-turbo OpenAI Speech to text Speech recognition model Audio input MIT Transcription, meeting notes, subtitles, translation to English

Pricing

All models are billed per token in AUD. Current per-model rates are listed in our model catalog.

Sign up

About the models

Z.ai GLM 5.2

For hard problems where reasoning quality justifies a higher cost per query.

Capabilities

  • Long-horizon reasoning with adjustable effort

  • Advanced coding, agentic engineering and tool use

  • Mathematical reasoning and problem-solving

  • Native, high-fidelity English and Mandarin, suited to organisations operating across APAC

  • FP8 weights with sparse attention for efficient serving

Use cases:

Chat, reasoning, code generation, agentic tasks, tool use, long-context analysis and bilingual tasks.

Gemma 4 26B A4B Instruct

Delivers near-dense quality at a fraction of the inference cost. Structured thinking mode, and a built-in function-call protocol. Accepts text, images and video.

Capabilities:

  • MoE architecture (26B total / ~4B active per token)

  • Vision input: up to 4 images per request

  • Video input: 1 clip per request, up to 60 seconds at 1 fps

  • Structured thinking mode for reasoning

  • Function calling and tool use

  • Multilingual support

  • Instruction-tuned for chat and tool-augmented workflows

Use cases:

Chat, reasoning, tool use, code generation, long-context analysis, multilingual support, question answering, image analysis, video analysis and document understanding.

Qwen3 Coder 30B A3B Instruct

Purpose-built for agentic and repository-level coding workflows. No thinking-mode tokens - direct response generation.

Capabilities:

  • Agentic coding at repository scale

  • Function calling and tool use (qwen3_coder parser format)

  • Multi-file refactoring, test generation and debugging

  • MoE architecture for fast decode (~3B parameters active per token)

Use cases:

Code generation, code completion, agentic coding, repository-level tasks, test generation, debugging, refactoring and tool use.

Whisper Large V3 Turbo

OpenAI Whisper Large V3 Turbo - faster variant of Whisper Large V3 with near-identical quality. Served via speaches (faster-whisper/CTranslate2). Supports MP3, WAV, FLAC, OGG, M4A.

Capabilities:

  • Automatic speech recognition in 99+ languages

  • Automatic language detection (language=multi)

  • Translation to English

  • MP3, WAV, FLAC, OGG/Opus and M4A input

  • Faster inference than Large V3 with minimal quality loss

Use cases:

Transcription, speech-to-text, meeting notes, subtitles, real-time transcription and translation.

Frequently asked questions

  • All inference runs on ResetData's sovereign infrastructure in Australia. Data stays onshore and is never used to train models.

  • Yes. Every model is served through OpenAI-compatible APIs, so tools and code built for the OpenAI API work by changing the base URL, API key and model ID. There's no migration. You can test any model in the Playground first, with no code needed.

  • Yes. This page lists every model currently live on Sovereign AI Inference and is updated as new models launch.

  • API keys can be IP-locked, so they only work from addresses you approve. You can manage keys, teams and projects in one place, with real-time dashboards showing usage per project.

  • Yes, ResetData also offers dedicated sovereign GPU infrastructure and GPUaaS. Talk to our team about dedicated capacity https://resetdata.com.au/contact

  • No. ResetData has a zero data retention policy, requests are processed in memory and not stored, and data is never used for model training.

Ready to deploy?