Reference
Noema models
Every released Noema model with its repository, size, base model, formats, license and release date.
NoemaAI post-trains small language models from Qwen bases and publishes them on Hugging Face (opens in a new tab). This reference is generated from the same data as the model pages.
On this page
Released models
| Model | Repository | Size | Base model | Formats | Released |
|---|---|---|---|---|---|
| Noema 1.5 4B Deep | Noema-1.5-4B-Deep (opens in a new tab) | 4B | Qwen3.5-4B | Transformers, GGUF, MLX (BF16, 8-bit, 6-bit, 4-bit) | Oct 13, 2026 |
| Noema 1.5 4B Fast | Noema-1.5-4B-Fast (opens in a new tab) | 4B | Qwen3.5-4B | Transformers, GGUF, MLX (BF16, 8-bit, 6-bit, 4-bit) | Oct 13, 2026 |
| Noema 1.5 2B | Noema-1.5-2B (opens in a new tab) | 2B | Noema 2B, trained from Qwen3.5-2B | Transformers, GGUF, MLX (BF16, 8-bit, 6-bit, 4-bit) | Aug 12, 2026 |
| Noema 2B | Noema-2B (opens in a new tab) | 2B | Qwen3.5-2B | Transformers, GGUF | Jul 6, 2026 |
Noema 2B is deprecated; use Noema 1.5 2B instead. Its weights stay available.
Every Noema release is open weights, not open source. See Model licenses for the terms of each release.
Context, modes and formats
| Model | Context | Modes | Formats |
|---|---|---|---|
| Noema 1.5 4B | 262,144 tokens (architecture limit); outputs evaluated to 81,920 | Thinking (default) · non-thinking, selected through the chat template | SafeTensors (BF16) · GGUF (BF16 to Q4_K_M) · MLX (BF16 to 4-bit) |
| Noema 1.5 2B | 262,144 tokens (architecture limit); independently validated to 24,576 | Non-thinking (recommended) · thinking | SafeTensors (BF16) · GGUF (BF16 to Q4_K_M) · MLX (BF16 to 4-bit) |
| Noema 2B | 262,144 tokens (architecture limit) | Non-thinking (trained and evaluated); thinking mode is not supported | SafeTensors (BF16) · GGUF (F16 to Q4_K_M) |
A model's context limit is not a promise that it fits in your device's memory. Size the context to the work and the hardware.
Repositories by format
Noema 2B has Transformers and GGUF repositories only. GGUF files are named after the repository and quantization, for example Noema-1.5-4B-Fast-Q4_K_M.gguf.
Lineage
Each model page credits the foundation model it was trained from.
- Noema 1.5 4B Deep: Qwen3.5-4B → Noema 1.5 4B Fast → Noema 1.5 4B Deep
- Noema 1.5 4B Fast: Qwen3.5-4B → Noema 1.5 4B Fast
- Noema 1.5 2B: Qwen3.5-2B → Noema 2B → Noema 1.5 2B
- Noema 2B: Qwen3.5-2B → Noema 2B