Tülu 3 70B

Overview

Allen AI's fully open RLHF stack on Llama 3.1 70B, beating Claude Haiku, GPT-3.5 Turbo, and GPT-4o-mini on standard reasoning and code benchmarks.

When to pick this model

• Self-hosted alternative to closed mid-tier APIs
• Math-heavy chat with GSM8K 93.5 territory performance
• Code assistance where HumanEval+ matters more than agentic loops
• Research projects that need a fully documented post-training pipeline
• Workloads that justify a 2x A100 footprint

VRAM requirements by quantization

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 70B spills past single consumer GPUs even at Q4_K_M (40 GB) — think dual-GPU or workstation cards. Stepping up to Q8_0 nearly doubles the footprint to 75 GB, and unquantized FP16 weights take 140 GB — plan your GPU around the Q4 or Q5 figure unless you specifically need the higher fidelity.

Without a GPU, Tülu 3 70B needs roughly 64 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 1 tokens/sec on entry-level GPUs, on the order of 6 tokens/sec on a mid-range card, and up to 20 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 70B 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.

Which GPU should you buy to run Tülu 3 70B?

To run Tülu 3 70B locally at Q4, you need ~40 GB of VRAM. The best value for this is a Apple Mac Studio (64+ GB unified memory).

Published benchmark scores

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

To put Tülu 3 70B in context: its GSM8K score of 93.5 ranks #2 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

• Beats Claude Haiku, GPT-3.5 Turbo, and GPT-4o-mini on key evals
• GSM8K 93.5 and HumanEval+ 92.4 at open weights
• Fully open SFT + DPO + RLVR recipe
• Strong instruction following and refusal calibration
• Stable, well-documented behavior for production deploys

Limitations

• ~40 GB VRAM at Q4 — needs serious hardware
• Bound by Llama 3.1 Community License
• No multimodal capabilities

Typical workloads

In our catalog grid, Tülu 3 70B is filed under Pro Chat, Reasoning — the use cases where its size/quality trade-off makes the most sense. Its tags translate to concrete workloads: multi-step reasoning and math-flavoured tasks.

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 70B · full Tülu recipe

Training: SFT + DPO + RLVR on 70B.

Quick start

ollama run tulu3:70b

Or 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 70B need?

At the recommended Q4_K_M quantization, Tülu 3 70B needs about 40 GB of VRAM. Q8_0 takes 75 GB, and unquantized FP16 weights take 140 GB.

Can Tülu 3 70B run without a GPU?

Yes — with roughly 64 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 70B support?

Tülu 3 70B supports a 125k-token context window (128,000 tokens).

Can I use Tülu 3 70B commercially?

Tülu 3 70B 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 70B on consumer hardware?

Our compatibility engine estimates on the order of 6 tokens/sec on a mid-range GPU and up to 20 tokens/sec on high-end cards, assuming the quantization fully fits in VRAM.

Which quantization of Tülu 3 70B should I download first?

Start with Q4_K_M (40 GB) — the standard size/quality sweet spot. Step up to Q5_K_M or Q8_0 only if you have VRAM headroom. It does not fit a single 24 GB consumer card — plan for multi-GPU or server hardware.