Models
Homegrown KiwiMate language models, built right here in New Zealand. We're just getting started — the family will grow over time.
KiwiMate Preview models
Our first generation of models, available today.
Our first model — small, fast, and proudly Kiwi.
KiwiMate-Mini-Preview is a compact 3B-parameter language model and the first release in the KiwiMate model family. It's an early preview build, so expect rough edges while we keep training and refining it.
- Parameters
- 3B
- Knowledge Cutoff
- Mid 2024 – Jan 2025
- Capabilities
- Text generation, Chat
- Pricing
- $1 input / $3 output per 1M tokens (NZD)
Pricing is indicative and not yet finalised — it may change before general availability.
View on Hugging FaceA bigger, multimodal model for more capable Kiwi intelligence.
KiwiMate-Small-Preview is a 10B-parameter multimodal language model that accepts image and text inputs and produces text output. It's an early preview build, so expect continued training and refinement.
- Parameters
- 10B
- Knowledge Cutoff
- January 2025
- Capabilities
- Image + text input, Text output
- Pricing
- $2 input / $5 output per 1M tokens (NZD)
Pricing is indicative and not yet finalised — it may change before general availability.
View on Hugging FaceOur first image generation model, now live on Hugging Face.
KiwiMate-image-Preview is KiwiMate's first text-to-image generation model, a LoRA adaptor fine-tuned on FLUX.2-klein-4B. It's an early preview build, so expect continued training and refinement.
- Parameters
- 4B
- Capabilities
- Text-to-image
- Pricing
- $0.10 per image (NZD)
KiwiMate 1.0 models
The next generation of KiwiMate models — now live.
The next generation of Mini — now multimodal.
KiwiMate-Mini-1.0 is the successor to KiwiMate-Mini-Preview: a 3B-parameter model that adds image understanding alongside text.
- Parameters
- 3B
- Context Length
- 128,000 tokens
- Capabilities
- Image + text input, Text output
- Pricing
- $1 input / $3 output per 1M tokens (NZD)
Pricing is indicative and not yet finalised — it may change before general availability.
⚠️ Fine-tuned from a Qwen Research-licensed base model, unlike the rest of the KiwiMate family (Apache 2.0) — commercial use terms differ.
View on Hugging FaceA leaner, faster follow-up to Small-Preview.
KiwiMate-Small-1.0 is a 9B-parameter model and the next iteration of the Small line, now with image understanding alongside text.
- Parameters
- 9B
- Context Length
- 262,144 tokens
- Capabilities
- Image + text input, Text output
- Pricing
- $2 input / $5 output per 1M tokens (NZD)
Pricing is indicative and not yet finalised — it may change before general availability.
View on Hugging FaceA larger, more capable model for demanding Kiwi intelligence.
KiwiMate-Medium-1.0 is a 27B-parameter multimodal model that accepts image and text inputs and produces text output.
- Parameters
- 27B
- Context Length
- 262,144 tokens
- Capabilities
- Image + text input, Text output
- Pricing
- $4 input / $8 output per 1M tokens (NZD)
Pricing is indicative and not yet finalised — it may change before general availability.
View on Hugging FaceOur flagship model — maximum capability for the most demanding tasks.
KiwiMate-Large-1.0 is a 36B-parameter multimodal model with 3B active parameters and the flagship of the 1.0 family, built for the most demanding Kiwi intelligence tasks.
- Parameters
- 36B
- Context Length
- 262,144 tokens
- Capabilities
- Image + text input, Text output
- Pricing
- $5 input / $10 output per 1M tokens (NZD)
Pricing is indicative and not yet finalised — it may change before general availability.
View on Hugging FaceMore models on the way
Larger and more capable KiwiMate models are in the works. Watch this space.