Aya 23 8B
8BCohere For AI
⚠ Superseded · Prefer Qwen3 8B Instruct
This index would suggest Qwen3 8B Instruct instead today — longer context (8K → 40K tokens). The numbers below are still accurate; they are just for a model you probably should not start with. Compare the two →
Multilingual specialist covering 23 languages. Strong for non-English local apps.
8K
Max Context
2
Quant Variants
GGUF Q4_K_M
Best Quality
96.8%
Accuracy Retained
Quantization Variants
Per-quant VRAM, quality loss, and inference speed on RTX 4090
Measured = site benchmarks · Estimated = formula · Community = public reports
Similar models
Compare with Llama 3.1Llama 3.1 8B Instruct
Meta Llama 3.1
Meta's flagship 8B model with 128K context. Best-in-class for local deployment.
Qwen2.5 7B Instruct
Alibaba Qwen2.5
Alibaba's highly optimized 7B. Punches well above its weight, especially in coding.
Phi-3.5 Mini Instruct
Microsoft Phi
Microsoft's tiny powerhouse. Best 4B model for on-device deployment.
Nous Hermes 3 Llama 3.1 8B
NousResearch
Fine-tuned Llama 3.1 8B with improved roleplay and instruction following.
Running Aya 23 8B locally
At Q4_K_M and 4K of context, Aya 23 8B needs about 5.6 GB — 4.6 GB of weights, 0.50 GB of KV cache and a 0.5 GB activation buffer. The smallest card in this index that clears that comfortably is the RTX 5060 Ti 8G at 8 GB, and 62 of the 63 cards here do. These are calculated figures, not measurements: the estimate stops counting a card as comfortable at 88% of its VRAM, which is roughly the room a desktop session needs.
What longer context costs
Going from 4K to 32K adds about 3.50 GB, taking the total to 9.5 GB. Weights do not move with context — only the KV cache does, and it grows linearly, so this is the number to watch when planning for long documents. The model's native window is 8K; holding all of it at Q4_K_M would need about 6 GB.
Which build to download
This index tracks 2 formats for it — GGUF, AWQ — across 2 levels. Q4_K_M carries the lowest published perplexity loss at 3.2%. The fastest level measured here is AWQ INT4 at 205 tok/s on an RTX 4090, batch 1. GGUF runs on llama.cpp and Ollama across NVIDIA, AMD and Apple silicon; AWQ and GPTQ target vLLM on CUDA and ROCm; EXL2 is ExLlamaV2 and CUDA only.
Common questions
- How much VRAM does Aya 23 8B need?
- About 5.6 GB at Q4_K_M with 4K of context and batch 1, rising to roughly 9.5 GB at 32K. That figure is weights plus KV cache plus a 10% activation buffer, calculated from the model's architecture rather than measured on a card.
- Will Aya 23 8B run on a 8GB GPU?
- Yes — at Q4_K_M and 4K context it needs about 5.6 GB, which leaves 2.4 GB spare on a RTX 5060 Ti 8G. That is the smallest card in this index that clears it comfortably; 62 of 63 do.
- Which quantization of Aya 23 8B should I use?
- Q4_K_M has the lowest published quality loss (3.2%), and Q4_K_M is the level most people run. All 2 levels in the index are Q4_K_M, AWQ INT4.
Where this model fits
Sized at Q4_K_M with a 4K context window, smallest card first. Comfortable means the estimate uses at most 88% of the memory.
- Runs comfortably on
- RTX 5060 Ti 8G8GB · +2.4RTX 50608GB · +2.4RTX 4060 Ti 8G8GB · +2.4RTX 40608GB · +2.4RTX 3070 Ti8GB · +2.4
- Tight but possible
- Mac M3 8G8GB
- Compared here
- GGUF vs AWQ
- Try it yourself
- Size it in the calculatorGenerate the run command
How to actually run this
Deployment guides for this model and this class of hardware.
This page's figures change when the model or the runtime does.Last updated 2026-09-21 RSS → /feed.xml