Tülu 3 8B
By Allen AI · United States
Updated 2026-07-13
Overview
Allen AI's fully open post-training recipe applied to Llama 3.1 8B, hitting 87.6 on GSM8K with all data, code, and evals released publicly.
When to pick this model
- Reproducible research on RLHF and DPO pipelines
- Drop-in replacement for Llama 3.1 8B Instruct with stronger math
- Instruction-following workloads needing high IFEval scores
- Teams that need to audit training data end-to-end
- Academic baselines requiring full provenance
VRAM requirements by quantization
| Quantization | VRAM required |
|---|---|
| Q4_K_M (recommended) | 6 GB |
| Q5_K_M | 7 GB |
| Q8_0 | 10 GB |
| FP16 (no quantization) | 16 GB |
VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.
In practice, Tülu 3 8B fits an 8 GB consumer card at Q4_K_M (6 GB). Stepping up to Q8_0 raises the footprint to 10 GB, and unquantized FP16 weights take 16 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.
Without a GPU, Tülu 3 8B needs roughly 10 GB of system RAM to run on CPU via llama.cpp or Ollama — workable for background jobs, but far slower than GPU inference. Throughput estimates from our compatibility engine: around 10 tokens/sec on entry-level GPUs, on the order of 30 tokens/sec on a mid-range card, and up to 80 tokens/sec on high-end hardware — assuming the chosen quantization fully fits in VRAM.
What hardware do you need
The table below matches Tülu 3 8B to common GPU memory tiers, using the highest-fidelity quantization that fully fits each card class. Spilling layers to system RAM works but costs most of the speed, so size your card to the quantization you actually want to run.
| GPU memory | Example cards | Best fit for Tülu 3 8B |
|---|---|---|
| 8 GB | RTX 5070 Laptop, RTX 5060, RTX 5060 Ti 8GB | Q5_K_M (7 GB used) |
| 12 GB | RTX 5070, RTX 5070 Ti Laptop, RTX 4080 Laptop | Q8_0 (10 GB used) |
| 16 GB | RTX 5080, RTX 4080 Super, Radeon RX 9070 XT | FP16 (16 GB used) |
| 24 GB | RTX 4090, Radeon RX 7900 XTX, RTX 5090 Laptop | FP16 (16 GB used) |
| 32 GB | RTX 5090 | FP16 (16 GB used) |
Which GPU should you buy to run Tülu 3 8B?
To run Tülu 3 8B locally at Q4, you need ~6 GB of VRAM. The best value for this is a RTX 5060 (8 GB VRAM).
As an Amazon Associate, BestLLMfor earns from qualifying purchases, at no extra cost to you. It does not influence our independent rankings.
Published benchmark scores
| Benchmark | Score |
|---|---|
| GSM8K | 87.6 |
| MATH | 42 |
| IFEval | 82.4 |
Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.
To put Tülu 3 8B in context: its GSM8K score of 87.6 ranks #4 of the 9 catalog models with a published GSM8K result (catalog median 83.1). Scores are author-published and measured under different harnesses, so treat gaps of a few points as noise rather than a verdict.
Strengths
- Best fully-open RLHF recipe shipped to date
- GSM8K 87.6 is class-leading at 8B
- IFEval 82.4 shows strong instruction adherence
- Training data, code, and evals all publicly available
- Stable behavior on standard chat benchmarks
Limitations
- Inherits the Llama 3.1 Community License
- No native vision or tool-use specialization
- Eclipsed at the frontier by larger open models
Typical workloads
In our catalog grid, Tülu 3 8B is filed under Chat, Instruction-following, Research — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: everyday chat, drafting and summarization.
The 125k-token context window covers long chats and mid-sized documents, though very large retrieval workloads will need chunking. It ships under the Llama 3.1 Community license — commercial use is generally possible but read the specific terms before embedding it in a product.
Architecture & training
Architecture: Dense Llama 3.1 8B · SFT + DPO + RLVR
Training: Public data + code + evals.
The reference open RLHF recipe at 8B — choose it when reproducibility and post-training transparency matter as much as benchmark scores.
Quick start
ollama run tulu3:8bOr use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.
Similar models worth comparing
Frequently asked questions
How much VRAM does Tülu 3 8B need?
At the recommended Q4_K_M quantization, Tülu 3 8B needs about 6 GB of VRAM. Q8_0 takes 10 GB, and unquantized FP16 weights take 16 GB.
Can Tülu 3 8B run without a GPU?
Yes — with roughly 10 GB of system RAM it runs CPU-only through llama.cpp or Ollama. Expect a fraction of GPU speed, which is fine for background or batch jobs but slow for interactive chat.
What context window does Tülu 3 8B support?
Tülu 3 8B supports a 125k-token context window (128,000 tokens).
Can I use Tülu 3 8B commercially?
Tülu 3 8B ships under the Llama 3.1 Community license. Commercial use is generally permitted subject to its terms — review the license text before shipping a product.
How fast is Tülu 3 8B on consumer hardware?
Our compatibility engine estimates on the order of 30 tokens/sec on a mid-range GPU and up to 80 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.
Which quantization of Tülu 3 8B should I download first?
Start with Q4_K_M (6 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It already fits an 8 GB card at Q5_K_M.