# Tileward model library: every model, with its size and its price

Our 35B-A3B build stays closer to the original model's answers than NVIDIA's NVFP4 build of the same model.

Page: https://tileward.com/model-library/

A served model is called by its id at `https://api.tileward.com/v1`, with any OpenAI-compatible client
([the docs](https://tileward.com/docs/) show how). Rates are list prices. Sizes are weights on disk, in decimal gigabytes. VRAM is
the memory a model needs to serve on one card, in decimal gigabytes, at what each entry says it was measured at.

## Every model

### Served today

Call by id now, billed per use.

- **Tileward 35B-A3B** (`Tileward-Qwen3.6-35B-A3B`). Compressed from Qwen3.6-35B-A3B. 35B parameters (3B active); context 262K; encoding TWX48 (Tileward); size 24.1 GB; VRAM 25.6 GB at 4,096 context; $0.12 input and $0.99 output per million tokens.
  Also answers to `Qwen3.6-35B-A3B-TW`, `tileward-35b-a3b`. [Results against the FP8 build](https://tileward.com/evals/#compressed-model). [The write-up](https://tileward.com/research/what-compression-inherits/).
- **Tileward Qwen3.8-27B** (`Tileward-Qwen3.8-27b`). Compressed from Qwen3.8-27B. 27B parameters, dense; context 262K; encoding TWFP4 (Tileward); size 20.1 GB; VRAM 35.0 GB at 262,144 context; $0.12 input and $0.96 output per million tokens.
  Licence: Apache-2.0.
- **Tileward PersonaPlex 7B** (`Tileward-PersonaPlex-7b`). Compressed from NVIDIA PersonaPlex. 7B parameters, dense; encoding TWX4 (Tileward); VRAM 18.1 GB at one conversation; $0.08 per minute.
  [The write-up](https://tileward.com/research/a-smaller-voice-model/).

### Ready to serve

Served on request, hosted or on your hardware.

- **Tileward GLM-5.3-Flash** (`Tileward-GLM-5.3-Flash`). Compressed from GLM-5.3-Flash. 321B parameters, mixture of experts; encoding TWX48 (Tileward); size 205.1 GB; price on request.
  Licence: MIT.
- **Tileward Nemotron 3 Nano 30B** (`Tileward-Nemotron-3-Nano-30B-A3B`). Compressed from Nemotron 3 Nano 30B-A3B. 30B parameters (3B active); size 21.1 GB; price on request.
  Licence: [NVIDIA Nemotron Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/).
- **Tileward Qwen3-Coder-30B-A3B** (`Tileward-Qwen3-Coder-30B-A3B`). Compressed from Qwen3-Coder-30B-A3B. 30B parameters (3B active); size 18.5 GB; price on request.
  Licence: Apache-2.0.
- **Tileward 35B-A3B TWX3** (`Tileward-Qwen3.6-35B-A3B-TWX3`). Compressed from Qwen3.6-35B-A3B. 35B parameters (3B active); encoding TWX3 (Tileward); size 20.1 GB; price on request.
  Licence: Apache-2.0.
- **Tileward 35B-A3B TWX2** (`Tileward-Qwen3.6-35B-A3B-TWX2`). Compressed from Qwen3.6-35B-A3B. 35B parameters (3B active); encoding TWX2 (Tileward); size 16.0 GB; price on request.
  Licence: Apache-2.0.
- **Tileward Kimi K3 TWC25** (`Tileward-Kimi-K3-TWC25`). Compressed from Kimi K3. 2.8T parameters (104B active); encoding TWC25 (Tileward); size 1,007.9 GB, 35% smaller than Kimi K3 as published (1.5×); VRAM 136.7 GB at weights on each of 8 cards; price on request.
  Licence: [Kimi K3 License](https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE).
- **Tileward Kimi K3 TWMX34** (`Tileward-Kimi-K3-TWMX34`). Compressed from Kimi K3. 2.8T parameters (104B active); encoding TWMX34 (Tileward); size 1,178.1 GB, 25% smaller than Kimi K3 as published (1.3×); VRAM 157.9 GB at weights on each of 8 cards; price on request.
  Licence: [Kimi K3 License](https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE).
- **Tileward VoiceChat 11B** (`Tileward-VoiceChat-11b`). Compressed from NVIDIA VoiceChat. 11B parameters, dense; encoding TWX4 (Tileward); VRAM 32 GB at one conversation; $0.25 per minute.
  Licence: [OpenMDW-1.1](https://huggingface.co/nvidia/NVIDIA-NemotronLabs-VoiceChat-11B/blob/main/LICENSE).
- **gpt-oss 20B** (`gpt-oss-20b`). Open weights, as published. 20B parameters (~3.6B active); context 8K; encoding MXFP4 (as shipped); size 13.7 GB; price on request.
  Licence: Apache-2.0.
- **gpt-oss 120B** (`gpt-oss-120b`). Open weights, as published. 117B parameters (~5.1B active); context 131K; encoding MXFP4 (as shipped); size 65.3 GB; price on request.
  Licence: Apache-2.0.
- **Qwen3.5-397B**. Compressed from Qwen3.5-397B. Price on request.
- **Qwen3-Next-80B-A3B**. Compressed from Qwen3-Next-80B-A3B. Price on request.

## A model that is not served today

We can serve it for you, hosted or on your hardware: [talk to an engineer](https://calendar.app.google/hg2DzFTxKJgwFWm19), or
[start free, no card](https://app.tileward.com/?join=explore).