FreeLLMAll models › CohereLabs
tiny-aya-earth
CohereLabs · license to accept · released 13 Feb 2026
Download No Arena votes
Arena score—Not voted on Arena yet
Size3.3 Bparameters
4-bit download2.0 GBweights only, Q4_K_M
Max context—
Where to use it for free
- Run it in one command with whichllm (free, open source):
uvx whichllm@latest plan "tiny-aya-earth"shows the GPU you need,uvx whichllm@latest run "tiny-aya-earth gguf"downloads it and starts a chat - Download and run it at home: official weights · GGUF files for Ollama and LM Studio (you accept the license on Hugging Face first)
No Arena votes yet
Arena has not voted on this model. It is listed because it is a popular small model on its lab's official Hugging Face page (6K downloads a month).
Speed on popular hardware
| Hardware | GPU memory | Fits? | Speed |
|---|---|---|---|
| RTX 3060 12 GB | 12 GB | Fits in GPU | 103 tok/s |
| RTX 4060 Ti 16 GB | 16 GB | Fits in GPU | 85 tok/s |
| RTX 3090 | 24 GB | Fits in GPU | 231 tok/s |
| RTX 4090 | 24 GB | Fits in GPU | 244 tok/s |
| RTX 5090 | 32 GB | Fits in GPU | 365 tok/s |
| RX 7900 XTX | 24 GB | Fits in GPU | 235 tok/s |
| 2 × RTX 3090 | 48 GB | Fits in GPU | 224 tok/s |
| 4 × RTX 3090 | 96 GB | Fits in GPU | 212 tok/s |
| Mac M4 Max 128 GB | 96 GB | Fits in GPU | 109 tok/s |
| Ryzen AI Max+ 395 128 GB | 96 GB | Fits in GPU | 76 tok/s |
| 8 × H100 SXM | 640 GB | Fits in GPU | 428 tok/s |
Estimate at 4-bit with 32K context, 64 GB of DDR5 for offload. Speed ≈ memory bandwidth × 78% ÷ data read per token, plus a fixed cost per token, calibrated on 44 real llama.cpp tests; typical error about ±10%. Compare all GPUs.
GPU memory needed
| Level | 8K context | 32K context | 128K context | Quality vs full |
|---|---|---|---|---|
| Q3_K_M | 2.4 GB | 2.7 GB | 4.0 GB | +10.5% noticeable loss |
| Q4_K_M | 2.7 GB | 3.0 GB | 4.3 GB | +2.8% small loss |
| Q5_K_M | 3.1 GB | 3.4 GB | 4.7 GB | +0.9% very small loss |
| Q6_K | 3.4 GB | 3.7 GB | 5.0 GB | +0.3% near-lossless |
| Q8_0 | 4.2 GB | 4.5 GB | 5.8 GB | ≈ 0% same as full |
| F16 | 7.3 GB | 7.6 GB | 8.9 GB | ≈ 0% same as full |
Weights + context cache (from the model's architecture) + runtime. Quality: increase in perplexity against the full model, from llama.cpp tests, scaled by model size. Simulate your own setup.
Share this page:XLinkedInRedditWhatsApp
Spotted a wrong number or a broken link? Report a mistake