Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Le Chat is fast (1,100 tok/s for flash queries on an updated Mistral Large). Download it at or

Mistral AI

200,628 subscribers

731,411 views • 1 year ago •via X (Twitter)

Science & Technology News & Politics Education

Anya Rossi• Live Now

Private livecam show

10 Comments

Quentin de Quelen1 year ago

I found faster, 1M tok/s cat /dev/random

CodeRabbit1 year ago

AI-first pull request reviewer with context-aware feedback, line-by-line code suggestions, and real-time chat.

mlukicic1 year ago

It is so fast that I was thinking that it is using Small 3. This is great!

Ossi K1 year ago

But im le tired :(

Quentin de Quelen1 year ago

So nice, go @MistralAI team! What about the relevancy of the result?

dune (Ø,G)1 year ago

Le Chat may be fast, but can it help me train my AI to fetch snacks? 🤔 With PublicAI, I can finally get rewarded for my snack-fetching data! 🍕💻

Qusai Ismael1 year ago

cant believe that we see others slow now, we should go back to type the code ourselves

0xblik 💎1 year ago

$LECHAT CA: 48B8EuNerXrwBah4MPpccF7u7jEto8L7HZcFcbU6pump

Silvio S.1 year ago

Woah

Mahdigh081 year ago

Le Chat may be fast, but with PublicAI, you can train your own AI while earning rewards! Who knew data could be so rewarding? 🐱💰 #DataIsTheNewGold

Related Videos

GLM 4.7 Flash is supported in mlx-lm 0.30.3 (h/t Ivan Fioravanti ᯅ) The 4-bit runs fast (43 tok/s generation, ~800 tok/s prefill) on a base M5 32GB laptop.

GLM 4.7 Flash is supported in mlx-lm 0.30.3 (h/t Ivan Fioravanti ᯅ) The 4-bit runs fast (43 tok/s generation, ~800 tok/s prefill) on a base M5 32GB laptop.

Awni Hannun

141,194 views • 6 months ago

I made a site to chat with Mistral AI's new 8x7B instruct model! - free + open source - streams at ~40 tokens/s - beats GPT-3.5 on most benchmarks Try it at

I made a site to chat with Mistral AI's new 8x7B instruct model! - free + open source - streams at ~40 tokens/s - beats GPT-3.5 on most benchmarks Try it at

Charlie Holtz

100,813 views • 2 years ago

BREAKING 🚨: Mistral is getting fully rebranded 🔥 - New logo! - Upgraded Le Chat UI - New perks on Le Chat - Upgraded news website - A new Android app 🔥 Get ready for Pro 🤖

🚨 AI News | TestingCatalog

53,759 views • 1 year ago

GLM-5.2 Nvidia NVFP4 No pruning, no quantising, exact same model card. 110 tok/s single stream. My friend is a genius, holy fudge. Here's the repo, he got full DS4-Flash on a 5090 + DDR5 at 38.5 tok/s This is a breakthrough.

GLM-5.2 Nvidia NVFP4 No pruning, no quantising, exact same model card. 110 tok/s single stream. My friend is a genius, holy fudge. Here's the repo, he got full DS4-Flash on a 5090 + DDR5 at 38.5 tok/s This is a breakthrough.

0xSero

156,611 views • 23 days ago

There is a Costco in LA known on tik tok as "Chinese Costco" or "Alhambra Costco" and it is notorious for being an awful experience.

There is a Costco in LA known on tik tok as "Chinese Costco" or "Alhambra Costco" and it is notorious for being an awful experience.

bumbadum

76,444 views • 1 year ago

Mistral AI is working on a new feature that allows users to set a Tone for their Le Chat conversations based on presets and custom prompts. A “personality” builder! 🔥

Mistral AI is working on a new feature that allows users to set a Tone for their Le Chat conversations based on presets and custom prompts. A “personality” builder! 🔥

TestingCatalog News 🗞

11,675 views • 10 months ago

Mistral Vibe 2.0 is now available on Le Chat Pro and Team plans. Build, maintain, and ship code faster with the terminal-native coding agent by Mistral AI. Here’s what’s new 🧵

Mistral Vibe 2.0 is now available on Le Chat Pro and Team plans. Build, maintain, and ship code faster with the terminal-native coding agent by Mistral AI. Here’s what’s new 🧵

Vibe

392,247 views • 6 months ago

That was fast: Google pushed an update to Gemini for Android last night that enabled auto sending queries when you invoke it via a swipe up gesture or long press of the power button. Previously queries only auto sent if you invoked Gemini via "Hey Google." Hopefully continued conversation is added next!

That was fast: Google pushed an update to Gemini for Android last night that enabled auto sending queries when you invoke it via a swipe up gesture or long press of the power button. Previously queries only auto sent if you invoked Gemini via "Hey Google." Hopefully continued conversation is added next!

Mishaal Rahman

64,870 views • 2 years ago

You buy this 1970's time capsule house, it has not been updated at all - do you enjoy it as is or renovate? 🤔

You buy this 1970's time capsule house, it has not been updated at all - do you enjoy it as is or renovate? 🤔

AlphaFox

15,012 views • 1 month ago

Introducing remote agents in Vibe and Mistral Medium 3.5. You can now launch remote agents in the cloud, including from the CLI or Le Chat. Plus, new Work mode in Le Chat for complex, multi-step tasks. 🧵

Introducing remote agents in Vibe and Mistral Medium 3.5. You can now launch remote agents in the cloud, including from the CLI or Le Chat. Plus, new Work mode in Le Chat for complex, multi-step tasks. 🧵

Vibe

194,542 views • 3 months ago

DFlash v0.1.4 : custom Metal verify kernels for quantized Qwen3 hybrid models, plus significant peak memory reduction at long context. M5 Max 40-core GPU, 64GB, stock mlx_lm baseline: Qwen3.6-35B-A3B-4bit: ► @ 1024 · 138.3 → 300.3 tok/s (2.20x) ► @ 2048 · 135.6 → 246.4 tok/s (1.81x) ► @ 4096 · 134.5 → 208.4 tok/s (1.56x) ► @ 8192 · 133.2 → 177.4 tok/s (1.33x) Qwen3.5-27B-4bit: ► @ 1024 · 33.5 → 79.0 tok/s (2.37x) ► @ 2048 · 33.1 → 70.2 tok/s (2.12x) ► @ 4096 · 31.5 → 55.7 tok/s (1.77x) ► @ 8192 · 33.9 → 45.3 tok/s (1.34x) Working on making this usable for agentic workloads goal is to never drop below baseline at any context depth. LLM decode is memory-bandwidth bound. M5 Max runs at 614 GB/s, that's 1.5x more than M1-M4 Max (400-410 GB/s). Results will vary on lower bandwidth chips.

DFlash v0.1.4 : custom Metal verify kernels for quantized Qwen3 hybrid models, plus significant peak memory reduction at long context. M5 Max 40-core GPU, 64GB, stock mlx_lm baseline: Qwen3.6-35B-A3B-4bit: ► @ 1024 · 138.3 → 300.3 tok/s (2.20x) ► @ 2048 · 135.6 → 246.4 tok/s (1.81x) ► @ 4096 · 134.5 → 208.4 tok/s (1.56x) ► @ 8192 · 133.2 → 177.4 tok/s (1.33x) Qwen3.5-27B-4bit: ► @ 1024 · 33.5 → 79.0 tok/s (2.37x) ► @ 2048 · 33.1 → 70.2 tok/s (2.12x) ► @ 4096 · 31.5 → 55.7 tok/s (1.77x) ► @ 8192 · 33.9 → 45.3 tok/s (1.34x) Working on making this usable for agentic workloads goal is to never drop below baseline at any context depth. LLM decode is memory-bandwidth bound. M5 Max runs at 614 GB/s, that's 1.5x more than M1-M4 Max (400-410 GB/s). Results will vary on lower bandwidth chips.

bstn 👁️

23,120 views • 3 months ago

What people don't realize is that a short '10 minute technical talk' is easily 35 slides worth of content and months of technical work. I've often struggled to use AI to generate slides (the lack of taste!) but two unlocks recently changed that: 1. Fast code generation → fast slides iteration 2. Writing my slides in HTML instead of generating graphics (models are better at writing code) Check out how I built my boilerplate with Mistral AI's Mistral Vibe. Available through the Le Chat Pro and team plan :) Get started with Mistral Vibe:

What people don't realize is that a short '10 minute technical talk' is easily 35 slides worth of content and months of technical work. I've often struggled to use AI to generate slides (the lack of taste!) but two unlocks recently changed that: 1. Fast code generation → fast slides iteration 2. Writing my slides in HTML instead of generating graphics (models are better at writing code) Check out how I built my boilerplate with Mistral AI's Mistral Vibe. Available through the Le Chat Pro and team plan :) Get started with Mistral Vibe:

Sarah Chieng

18,300 views • 3 months ago

The RTX 3090 is a 5-year-old GPU and it still runs a 27B model at 20 tok/s I tested Qwen3.5:27b across 3 generations of NVIDIA: 5090 → ~60 tok/s 4090 → ~40 tok/s 3090 → ~20 tok/s Perfectly linear scaling. Double the generation, double the speed.

The RTX 3090 is a 5-year-old GPU and it still runs a 27B model at 20 tok/s I tested Qwen3.5:27b across 3 generations of NVIDIA: 5090 → ~60 tok/s 4090 → ~40 tok/s 3090 → ~20 tok/s Perfectly linear scaling. Double the generation, double the speed.

stevibe

142,493 views • 4 months ago

Updated trailer for an updated game! Good Night Pan Pan is available NOW on Steam!

Updated trailer for an updated game! Good Night Pan Pan is available NOW on Steam!

SUZAKU

27,849 views • 1 year ago

Mistral OCR 3 now returns confidence scores, at page level or word level. Know exactly how sure the model is about what it extracted. Available now via API and to test in Mistral Studio.

Mistral OCR 3 now returns confidence scores, at page level or word level. Know exactly how sure the model is about what it extracted. Available now via API and to test in Mistral Studio.

Mistral AI for Developers

29,944 views • 3 months ago

Updated my Grok Rate Limit Display Chrome extension. Added Grok Imagine support. Now you can easily track how many chat queries, and imagine video and image generations you have left and how long you have to wait for more. Download. or just search the chrome web store for "Grok Rate Limit Display" note: Waiting for google to approve ver 1.6.0 with imagine support

Updated my Grok Rate Limit Display Chrome extension. Added Grok Imagine support. Now you can easily track how many chat queries, and imagine video and image generations you have left and how long you have to wait for more. Download. or just search the chrome web store for "Grok Rate Limit Display" note: Waiting for google to approve ver 1.6.0 with imagine support

️️️️ ️ᅠ‏️️️️ ️ᅠ️️️️ ️️️️️ ️ᅠ

69,461 views • 2 months ago

New short course with Mistral AI ! Mistral's open-source Mixtral 8x7B model uses a "mixture of experts" (MoE) architecture. Unlike a standard transformer, an MoE model has multiple expert feed-forward networks (8 in this case), with a gating network selecting two experts at inference time. This enables MoE to match the performance of a large model but faster inference. Mixtral 8x7B has 46.7B parameters but activates only 12.9B at inference to predict the next token. In Getting Started with Mistral, you’ll learn from Mistral’s Sophia Yang, Ph.D. to: - Explore Mistral's open-source models (Mistral 7B, Mixtral 8x7B) and commercial models via API calls and Mistral AI's Le Chat website - Implement JSON mode to generate structured outputs to integrate directly into larger software systems. - Use function calling for Tool Use, such as calling custom Python code that queries tabular data - Ground your LLM's response with external knowledge sources using RAG - Build a Mistral-powered chat interface that can reference external documents This course will help deepen your prompt engineering skills. Please sign up here:

New short course with Mistral AI ! Mistral's open-source Mixtral 8x7B model uses a "mixture of experts" (MoE) architecture. Unlike a standard transformer, an MoE model has multiple expert feed-forward networks (8 in this case), with a gating network selecting two experts at inference time. This enables MoE to match the performance of a large model but faster inference. Mixtral 8x7B has 46.7B parameters but activates only 12.9B at inference to predict the next token. In Getting Started with Mistral, you’ll learn from Mistral’s Sophia Yang, Ph.D. to: - Explore Mistral's open-source models (Mistral 7B, Mixtral 8x7B) and commercial models via API calls and Mistral AI's Le Chat website - Implement JSON mode to generate structured outputs to integrate directly into larger software systems. - Use function calling for Tool Use, such as calling custom Python code that queries tabular data - Ground your LLM's response with external knowledge sources using RAG - Build a Mistral-powered chat interface that can reference external documents This course will help deepen your prompt engineering skills. Please sign up here:

Andrew Ng

387,018 views • 2 years ago

First results are in. Llama 4 Maverick 17B active / 400B total is blazing fast with MLX on an M3 Ultra. Here is the 4-bit model generating 1100 tokens at 50 tok/sec:

First results are in. Llama 4 Maverick 17B active / 400B total is blazing fast with MLX on an M3 Ultra. Here is the 4-bit model generating 1100 tokens at 50 tok/sec:

Awni Hannun

149,855 views • 1 year ago

DeepSeek-v4 now runs at ~23-26 tok/s on MLX! I made some custom kernels for the sinkhorn and it took gen speeds for 17 -> 26 tok/s. The weights are also significantly smaller thanks to Pedro Cuenca tip about keeping the experts in MXFP4! Now you can use it to power your local coding agents (PI, Open code, Hermes agent or even CC) PR:

DeepSeek-v4 now runs at ~23-26 tok/s on MLX! I made some custom kernels for the sinkhorn and it took gen speeds for 17 -> 26 tok/s. The weights are also significantly smaller thanks to Pedro Cuenca tip about keeping the experts in MXFP4! Now you can use it to power your local coding agents (PI, Open code, Hermes agent or even CC) PR:

Prince Canuma

58,451 views • 3 months ago

Crazy Square ⚽️ Soccer! An action packed, fast paced game. Great for large classes. It is requested every year, and it was 3rd grades 1st time to play. To say it was a hit is an understatement! #mhepe #risdmoves #physed #risdbetheone #mhego4thegold

Crazy Square ⚽️ Soccer! An action packed, fast paced game. Great for large classes. It is requested every year, and it was 3rd grades 1st time to play. To say it was a hit is an understatement! #mhepe #risdmoves #physed #risdbetheone #mhego4thegold

Christina Fuller

13,109 views • 1 year ago