Blog

News, insight & more

News, insight & more

Learn more about Pruna, our approach and company news

Try Our Performance Models

Try Our OSS Models

Announcement

Featured

・

Introducing the Pruna Build Program

Discover the Pruna Build Program for teams that actually build and sca...

Bertrand Charpentier

・

Announcement

Featured

・

Introducing the Pruna Build Program

Discover the Pruna Build Program for teams that actually build and sca...

Bertrand Charpentier

・

Announcement

Featured

・

Introducing the Pruna Build Program

Discover the Pruna Build Program for teams that actually build and sca...

Bertrand Charpentier

・

Latest articles

All

News

Technical Articles and Guides

David

・

Technical Article

How Pruna Optimizes Models: Tracing the Smash Function and DAG Executi...

In this community blog, we will analyze how pruna works and optimizes...

David

・

Technical Article

How Pruna Optimizes Models: Tracing the Smash Function and DAG Executi...

In this community blog, we will analyze how pruna works and optimizes...

Bertrand Charpentier

・

Technical Article

Sustainable AI Starts with Efficient AI

Explore the sustainability impact of AI, including how to measure it a...

Bertrand Charpentier

・

Technical Article

Sustainable AI Starts with Efficient AI

Explore the sustainability impact of AI, including how to measure it a...

Bertrand Charpentier

・

Announcement

Introducing the Pruna Build Program

Discover the Pruna Build Program for teams that actually build and sca...

Bertrand Charpentier

・

Announcement

Introducing the Pruna Build Program

Discover the Pruna Build Program for teams that actually build and sca...

Bertrand Charpentier

・

Announcement

First Prune: Celebrating One Year of Pruna OSS

Join us for First Prune as we celebrate one year of Pruna OSS and look...

Bertrand Charpentier

・

Announcement

First Prune: Celebrating One Year of Pruna OSS

Join us for First Prune as we celebrate one year of Pruna OSS and look...

Quentin Sinig

・

Performance Models

How Looties Built an AI-Powered Marketplace with Pruna

How Looties uses AI to speed up marketplace listings with premium imag...

Quentin Sinig

・

Performance Models

How Looties Built an AI-Powered Marketplace with Pruna

How Looties uses AI to speed up marketplace listings with premium imag...

Bertrand Charpentier

・

Release Note

Pruna 0.3.2: More OSS Algos, More Ways to Optimize

Explore the latest updates in Pruna 0.3.2, including new algorithms, t...

Bertrand Charpentier

・

Release Note

Pruna 0.3.2: More OSS Algos, More Ways to Optimize

Explore the latest updates in Pruna 0.3.2, including new algorithms, t...

John Rachwan
Quentin Sinig

・

Announcement

Introducing P-Video: Fast and Cheap video generation

P-Video, the fastest and cheapest video generation on the market.

John Rachwan
Quentin Sinig

・

Announcement

Introducing P-Video: Fast and Cheap video generation

P-Video, the fastest and cheapest video generation on the market.

John Rachwan
David
Bertrand Charpentier

・

Announcement

Accelerating FLUX.2 [flex]: Now Design x3 Faster

FLUX.2 [flex] achieves x3 speedups

John Rachwan
David
Bertrand Charpentier

・

Announcement

Accelerating FLUX.2 [flex]: Now Design x3 Faster

FLUX.2 [flex] achieves x3 speedups

Bertrand Charpentier

・

Announcement

FLUX.2 [flex] Design Challenge

Read the rules to participate in the FLUX.2 [flex] Design Challenge.

Bertrand Charpentier

・

Announcement

FLUX.2 [flex] Design Challenge

Read the rules to participate in the FLUX.2 [flex] Design Challenge.

Quentin Sinig

・

Performance Models

How Fimo.ai Built an Image Gen Feature with Pruna.ai

Launch case study on our early collaboration with Strapi around Fimo.A...

Quentin Sinig

・

Performance Models

How Fimo.ai Built an Image Gen Feature with Pruna.ai

Launch case study on our early collaboration with Strapi around Fimo.A...

Bertrand Charpentier

・

Announcement

The 2025 Pruna’s Recap

The wins, lessons and highlights of Pruna in 2025

Bertrand Charpentier

・

Announcement

The 2025 Pruna’s Recap

The wins, lessons and highlights of Pruna in 2025

Quentin Sinig

・

Performance Models

P-Image and P-Image Edit are live on Scenario

P-Image and P-Image Edit are fast, affordable image generation and edi...

Quentin Sinig

・

Performance Models

P-Image and P-Image Edit are live on Scenario

P-Image and P-Image Edit are fast, affordable image generation and edi...

Bertrand Charpentier

・

Announcement

Everything You Need to Know about the Flux 2 x Pruna Contest

Flux 2 x Pruna Contest overview and details.

Bertrand Charpentier

・

Announcement

Everything You Need to Know about the Flux 2 x Pruna Contest

Flux 2 x Pruna Contest overview and details.

John Rachwan
Louis Leconte
David
Bertrand Charpentier

・

Announcement

Flux 2: A 110GB Model Now Running on One H100 and 4.2× Faster!

Flux 2 running on one H100 and 4.2x faster.

John Rachwan
Louis Leconte
David
Bertrand Charpentier

・

Announcement

Flux 2: A 110GB Model Now Running on One H100 and 4.2× Faster!

Flux 2 running on one H100 and 4.2x faster.

Bertrand Charpentier

・

Technical Article

Effective Prompting for Generative Vision Models

A prompting guide for generative vision models.

Bertrand Charpentier

・

Technical Article

Effective Prompting for Generative Vision Models

A prompting guide for generative vision models.

Quentin Sinig
John Rachwan

・

Performance Models

Inside Story: The Pruna × Replicate Recipe for the Fastest Inference

This is the origin story of the Pruna × Replicate partnership. From fi...

Quentin Sinig
John Rachwan

・

Performance Models

Inside Story: The Pruna × Replicate Recipe for the Fastest Inference

This is the origin story of the Pruna × Replicate partnership. From fi...

Louis Leconte
Bertrand Charpentier

・

Technical Article

vLLM Supercharged: Pruna’s Quantization Boost on H100

Discover how to deliver faster inference with vLLM and Pruna by integr...

Louis Leconte
Bertrand Charpentier

・

Technical Article

vLLM Supercharged: Pruna’s Quantization Boost on H100

Discover how to deliver faster inference with vLLM and Pruna by integr...

Bertrand Charpentier

・

Technical Article

An Introduction to the Architectures Powering the Current LLMs

Explore the core intuitions behind the current LLM architectures

Bertrand Charpentier

・

Technical Article

An Introduction to the Architectures Powering the Current LLMs

Explore the core intuitions behind the current LLM architectures

John Rachwan
Bertrand Charpentier

・

Technical Article

Slashing torch.compile Warmup & LoRA Swapping Times with Pruna

Learn how to achieve zero warmup using torch.compile with Pruna and cu...

John Rachwan
Bertrand Charpentier

・

Technical Article

Slashing torch.compile Warmup & LoRA Swapping Times with Pruna

Learn how to achieve zero warmup using torch.compile with Pruna and cu...

John Rachwan
Bertrand Charpentier

・

Announcement

Wan 2.2 Image Juiced: The Next Level for Efficient 2 Megapixels Image...

Wan 2.2. Image is now the new cheapest and fastest model to generate 2...

John Rachwan
Bertrand Charpentier

・

Announcement

Wan 2.2 Image Juiced: The Next Level for Efficient 2 Megapixels Image...

Wan 2.2. Image is now the new cheapest and fastest model to generate 2...

John Rachwan
Bertrand Charpentier
David

・

Announcement

Wan 2.2 Video Juiced: The Fastest & Cheapest Video Generation!

We optimized it to make it the fastest and cheapest video generation m...

John Rachwan
Bertrand Charpentier
David

・

Announcement

Wan 2.2 Video Juiced: The Fastest & Cheapest Video Generation!

We optimized it to make it the fastest and cheapest video generation m...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

Efficient Development, Efficient Production: ROAST Deploys 4.5x Faster...

ROAST used Pruna to accelerate image generation for their AI-powered d...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

Efficient Development, Efficient Production: ROAST Deploys 4.5x Faster...

ROAST used Pruna to accelerate image generation for their AI-powered d...

John Rachwan
Bertrand Charpentier
David

・

Announcement

Wan Image Juiced: The Fastest 2 Megapixels Image Generation Model

Wan 2.1 is the state-of-the-art model for generating high-quality vide...

John Rachwan
Bertrand Charpentier
David

・

Announcement

Wan Image Juiced: The Fastest 2 Megapixels Image Generation Model

Wan 2.1 is the state-of-the-art model for generating high-quality vide...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

Pruna Cuts GLiNER Latency by x2 for the Largest Cloud Monitoring Platf...

A story about speed, savings, and what happens when you squeeze every...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

Pruna Cuts GLiNER Latency by x2 for the Largest Cloud Monitoring Platf...

A story about speed, savings, and what happens when you squeeze every...

John Rachwan
Bertrand Charpentier
David

・

Announcement

Flux-Kontext-Juiced: State-of-the-art Image Editing 5x Faster!

Flux-Kontext is now open-source! We compressed it, deployed it, evalua...

John Rachwan
Bertrand Charpentier
David

・

Announcement

Flux-Kontext-Juiced: State-of-the-art Image Editing 5x Faster!

Flux-Kontext is now open-source! We compressed it, deployed it, evalua...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

How Pruna Delivered Up to 3.6x Faster Inference for BRIA 3.2

How we helped BRIA achieve up to 3.6× faster inference for their open-...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

How Pruna Delivered Up to 3.6x Faster Inference for BRIA 3.2

How we helped BRIA achieve up to 3.6× faster inference for their open-...

Bertrand Charpentier

・

Technical Article

SOTA Optimization for Image-Generation with Pruna Open Source

Pruna sets new standards for compressing image-generation models in th...

Bertrand Charpentier

・

Technical Article

SOTA Optimization for Image-Generation with Pruna Open Source

Pruna sets new standards for compressing image-generation models in th...

Louis Leconte
Quentin Sinig
Bertrand Charpentier

・

Technical Article

Llama-Juiced: The Fastest Llama 3.1-8B Endpoint (1.89× Faster than vLL...

Discover how we optimized inference latency on Llama 3.1-8B using Prun...

Louis Leconte
Quentin Sinig
Bertrand Charpentier

・

Technical Article

Llama-Juiced: The Fastest Llama 3.1-8B Endpoint (1.89× Faster than vLL...

Discover how we optimized inference latency on Llama 3.1-8B using Prun...

Stephan Günnemann
Bertrand Charpentier
David

・

Technical Article

Measuring What Matters: Objective Metrics for Image Generation Assessm...

Generating high-quality visuals with state-of-the-art models is now mo...

Stephan Günnemann
Bertrand Charpentier
David

・

Technical Article

Measuring What Matters: Objective Metrics for Image Generation Assessm...

Generating high-quality visuals with state-of-the-art models is now mo...

Bertrand Charpentier

・

Technical Article

Optimization Agent: Automatically Finding the Best AI Compression!

When it comes to scaling AI efficiently, configuration choices can mak...

Bertrand Charpentier

・

Technical Article

Optimization Agent: Automatically Finding the Best AI Compression!

When it comes to scaling AI efficiently, configuration choices can mak...

alisdair broshar

・

Announcement

5x Faster Inference Speeds on Serverless GPUs with Pruna AI and Koyeb

By combining Pruna with Koyeb, you can speed up your model optimizatio...

alisdair broshar

・

Announcement

5x Faster Inference Speeds on Serverless GPUs with Pruna AI and Koyeb

By combining Pruna with Koyeb, you can speed up your model optimizatio...

flux-juiced the best Flux optimization out there
David
John Rachwan
Bertrand Charpentier

・

Announcement

FLUX-Juiced: The Fastest Image Generation Endpoint (2.6x Faster!)

While compression is a widely applied technique that can reduce infere...

flux-juiced the best Flux optimization out there
David
John Rachwan
Bertrand Charpentier

・

Announcement

FLUX-Juiced: The Fastest Image Generation Endpoint (2.6x Faster!)

While compression is a widely applied technique that can reduce infere...

An overview of AI optimization Techniques
David
Bertrand Charpentier

・

Technical Article

An Introduction to AI Model Optimization Techniques

We will go over high level speed/cost/quality impact for AI optimizati...

An overview of AI optimization Techniques
David
Bertrand Charpentier

・

Technical Article

An Introduction to AI Model Optimization Techniques

We will go over high level speed/cost/quality impact for AI optimizati...

Louis Leconte
Bertrand Charpentier

・

Technical Article

Quantization for Image Generation Models to 3 bits: Shrinking Models,...

Quantization isn’t just a technical optimization—it’s a game-changer f...

Louis Leconte
Bertrand Charpentier

・

Technical Article

Quantization for Image Generation Models to 3 bits: Shrinking Models,...

Quantization isn’t just a technical optimization—it’s a game-changer f...

John Rachwan
Bertrand Charpentier
Quentin Sinig

・

Technical Article

Accelerating Image Generation: Going Beyond API Optimization from FAL,...

We compare the FLUX APIs provided by Replicate, fal, together.ai, and...

John Rachwan
Bertrand Charpentier
Quentin Sinig

・

Technical Article

Accelerating Image Generation: Going Beyond API Optimization from FAL,...

We compare the FLUX APIs provided by Replicate, fal, together.ai, and...

Bertrand Charpentier

・

Release Note

Auto-Caching: Pushing the Limits of Caching for FLUX

We introduce Auto-Caching, a novel caching algorithm that achieves sta...

Bertrand Charpentier

・

Release Note

Auto-Caching: Pushing the Limits of Caching for FLUX

We introduce Auto-Caching, a novel caching algorithm that achieves sta...

Bertrand Charpentier
John Rachwan
Rayan Nait Mazi
Stephan Günnemann

・

Announcement

Pruna AI Open Sources Its AI Optimization Engine

Everyone Can Make AI Models Faster, Smaller, Cheaper, Greener

Bertrand Charpentier
John Rachwan
Rayan Nait Mazi
Stephan Günnemann

・

Announcement

Pruna AI Open Sources Its AI Optimization Engine

Everyone Can Make AI Models Faster, Smaller, Cheaper, Greener

Amine Hamouda
Bertrand Charpentier

・

Technical Article

Quantized LLMs on edge devices: Techniques and Challenges

Amine Hamouda
Bertrand Charpentier

・

Technical Article

Quantized LLMs on edge devices: Techniques and Challenges

toongether & pruna
John Rachwan
Bertrand Charpentier

・

Performance Models

Beating torch.compile: 12× faster compilation time at Toongether.ai

Toongether cut latency by 40%, eliminated the 10-minute warm-up bottle...

toongether & pruna
John Rachwan
Bertrand Charpentier

・

Performance Models

Beating torch.compile: 12× faster compilation time at Toongether.ai

Toongether cut latency by 40%, eliminated the 10-minute warm-up bottle...

deepseek
Bertrand Charpentier
Quentin Sinig

・

Leadership Thought

DeepSeek: an opportunity for Efficient AI

DeepSeek, an AI company, demonstrates efficient training of large lang...

deepseek
Bertrand Charpentier
Quentin Sinig

・

Leadership Thought

DeepSeek: an opportunity for Efficient AI

DeepSeek, an AI company, demonstrates efficient training of large lang...

comparison-flux-torch-compile-tensor-rt-pruna
John Rachwan
Bertrand Charpentier

・

Technical Article

Unlock 2.7x Speedup for FLUX.1[dev]: A Comparison between torch.compil...

FLUX.1[dev] image generation can be optimized using torch.compile for...

comparison-flux-torch-compile-tensor-rt-pruna
John Rachwan
Bertrand Charpentier

・

Technical Article

Unlock 2.7x Speedup for FLUX.1[dev]: A Comparison between torch.compil...

FLUX.1[dev] image generation can be optimized using torch.compile for...

Quentin Sinig
John Rachwan

・

Performance Models

From 0 to 100% Faster: How Scenario Optimized Image Generation in 3 We...

That moment came when Scenario explored a new use case: an image upsca...

Quentin Sinig
John Rachwan

・

Performance Models

From 0 to 100% Faster: How Scenario Optimized Image Generation in 3 We...

That moment came when Scenario explored a new use case: an image upsca...

John Rachwan
Bertrand Charpentier

・

Technical Article

Pruna + Triton: A Winning Combination for High-Performance AI Deployme...

Combining Pruna's optimization techniques with NVIDIA's Triton Inferen...

John Rachwan
Bertrand Charpentier

・

Technical Article

Pruna + Triton: A Winning Combination for High-Performance AI Deployme...

Combining Pruna's optimization techniques with NVIDIA's Triton Inferen...

ml-ops-pruna
Bertrand Charpentier

・

Technical Article

Standard Model Compression in ML Pipeline

Integrating model compression techniques like pruning and quantization...

ml-ops-pruna
Bertrand Charpentier

・

Technical Article

Standard Model Compression in ML Pipeline

Integrating model compression techniques like pruning and quantization...

John Rachwan
Bertrand Charpentier

・

Technical Article

Boost Your Replicate Models with Pruna AI: A Step-by-Step Guide

Learn to optimize deep learning models using Pruna for faster inferenc...

John Rachwan
Bertrand Charpentier

・

Technical Article

Boost Your Replicate Models with Pruna AI: A Step-by-Step Guide

Learn to optimize deep learning models using Pruna for faster inferenc...

Quentin Sinig

・

Performance Models

Solving Evaluation for 15 AI Models at a German Automotive Giant

Evaluation is a complex topic, as it involves numerous variables to co...

Quentin Sinig

・

Performance Models

Solving Evaluation for 15 AI Models at a German Automotive Giant

Evaluation is a complex topic, as it involves numerous variables to co...

・

Technical Article

Hugging Face & Pruna AI: SmolLM2, Now 7x Smaller and 2x Faster

Pruna AI collaborated with Hugging Face to push the limits of model ef...

・

Technical Article

Hugging Face & Pruna AI: SmolLM2, Now 7x Smaller and 2x Faster

Pruna AI collaborated with Hugging Face to push the limits of model ef...

Bertrand Charpentier
Quentin Sinig

・

Announcement

Pruna AI has 9 papers at NeurIPS 2024

We are excited to announce that the Pruna AI team will present 9 paper...

Bertrand Charpentier
Quentin Sinig

・

Announcement

Pruna AI has 9 papers at NeurIPS 2024

We are excited to announce that the Pruna AI team will present 9 paper...

Bertrand Charpentier

・

Technical Article

Deep Dive Into Flux: Everything You Need to Know!

Deep dive into flux: our research pov, performance benchmarks, and eve...

Bertrand Charpentier

・

Technical Article

Deep Dive Into Flux: Everything You Need to Know!

Deep dive into flux: our research pov, performance benchmarks, and eve...

Bertrand Charpentier
Quentin Sinig

・

Announcement

Pruna AI secures $6.5 million to optimize machine learning models and...

Pruna AI, the AI optimization engine empowering ML teams to simplify c...

Bertrand Charpentier
Quentin Sinig

・

Announcement

Pruna AI secures $6.5 million to optimize machine learning models and...

Pruna AI, the AI optimization engine empowering ML teams to simplify c...

Quentin Sinig

・

Leadership Thought

How Pruna AI Contributes in Shaping France’s National AI Strategy?

🇪🇺 Today’s, we are incredibly proud to share Pruna AI's role in shap...

Quentin Sinig

・

Leadership Thought

How Pruna AI Contributes in Shaping France’s National AI Strategy?

🇪🇺 Today’s, we are incredibly proud to share Pruna AI's role in shap...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

Flux Schnell 3x Faster, 3x Cheaper? Mission Accomplished for Each AI

Each AI recently reached out to Pruna AI with a straightforward goal:...

Quentin Sinig
Bertrand Charpentier

・

Performance Models

Flux Schnell 3x Faster, 3x Cheaper? Mission Accomplished for Each AI

Each AI recently reached out to Pruna AI with a straightforward goal:...

Rayan Nait Mazi
Quentin Sinig

・

Leadership Thought

Behind The Scenes of the European AI Program

On October 1st, we officially joined Meta’s European AI Program. While...

Rayan Nait Mazi
Quentin Sinig

・

Leadership Thought

Behind The Scenes of the European AI Program

On October 1st, we officially joined Meta’s European AI Program. While...

Quentin Sinig

・

Technical Article

Achieve 2x to 4x Efficiency Gains on Databricks DBRX with Pruna AI

Check how pruning 4-bit quantized versions of DBRX-Base & DBRX-Instruc...

Quentin Sinig

・

Technical Article

Achieve 2x to 4x Efficiency Gains on Databricks DBRX with Pruna AI

Check how pruning 4-bit quantized versions of DBRX-Base & DBRX-Instruc...

Quentin Sinig

・

Technical Article

Revolutionizing Protein Structure Prediction The Power of Advanced ML...

As ML models take on increasingly complex biological challenges, the i...

Quentin Sinig

・

Technical Article

Revolutionizing Protein Structure Prediction The Power of Advanced ML...

As ML models take on increasingly complex biological challenges, the i...

cover pruna brand identity
Bertrand Charpentier
Quentin Sinig

・

Announcement

print("Hello, world! Pruna’s New Brand Identity & Editorial Line is He...

We’re excited to kick off our blog with this first article, unveiling...

cover pruna brand identity
Bertrand Charpentier
Quentin Sinig

・

Announcement

print("Hello, world! Pruna’s New Brand Identity & Editorial Line is He...

We’re excited to kick off our blog with this first article, unveiling...

Performance models are faster, cheaper, smaller, and greener.

Inefficient models drive up costs, slow down your productivity and increase carbon emissions. With Pruna, make your AI more accessible and sustainable.

Pruna models are faster, cheaper, smaller, and greener.

Inefficient models drive up costs, slow down your productivity and increase carbon emissions. With Pruna, make your AI more accessible and sustainable.

Pruna models are faster, cheaper, smaller, and greener.

Inefficient models drive up costs, slow down your productivity and increase carbon emissions. With Pruna, make your AI more accessible and sustainable.

Pruna

P-Image-Try-On

P-Video-Replace

P-Video-Animate

P-Video-Avatar

P-Image-Upscale

P-Video

P-Image

P-Image-Edit

All models

Pricing

Resources

Docs

Demos

Blog

InferBench

Company

About us

Contact

Careers

Logo kit

Join our Discord

Meet Pruna at events

Built with Pretzels & Croissants 🥨 🥐

@2026 PrunaAI

Terms

Privacy Notice

Legal Notice

Pruna

P-Image-Try-On

P-Video-Replace

P-Video-Animate

P-Video-Avatar

P-Image-Upscale

P-Video

P-Image

P-Image-Edit

All models

Pricing

Resources

Docs

Demos

Blog

InferBench

Company

About us

Contact

Careers

Logo kit

Join our Discord

Meet Pruna at events

Built with Pretzels & Croissants 🥨 🥐

Terms

Privacy Notice

Legal Notice

@2026 PrunaAI

Pruna

P-Image-Try-On

P-Video-Replace

P-Video-Animate

P-Video-Avatar

P-Image-Upscale

P-Video

P-Image

P-Image-Edit

All models

Pricing

Resources

Docs

Demos

Blog

InferBench

Company

About us

Contact

Careers

Logo kit

Join our Discord

Meet Pruna at events

Built with Pretzels & Croissants 🥨 🥐

@2026 PrunaAI

Terms

Privacy Notice

Legal Notice