Noctus
@noctus91
Benchmarking for fun | Ambassador @MistralAI |
Open to interesting projects & collaborations 📩
361 Following 1.4K Followers
Can an STT model make sense of a 135 year old recording?
I tested @MistralAI new Voxtral transcription model on a noisy French phonograph recording of Gustave Eiffel from 1891.
It matched 100 of 107 words in INA’s published excerpts (6.5% WER), while transcribing 75 additional words omitted from the reference.
It handled Eiffel’s dictation well but struggled with proper names and some guest dialogue.
A fascinating test of modern speech recognition on audio from 1891.
Show more
A 9B model challenging much larger AI models at decision making.
@MicrosoftAI Decision-1 is post-trained on Qwen3.5-9B and reportedly delivers up to 35× faster inference while outperforming larger models on decision benchmarks.
Show more
Introducing Microsoft-Decision-1, our new model for fast decision-making.
It delivers top performance on structured decision tasks, outperforming both LLMs and other decision models in latency and quality.
We’re already testing it across Microsoft for everything from incident response and quality control to scientific discovery.
Show more
The world’s richest man is literally having a total meltdown.
Dear Prime Minister Ambani,
Please accept my humble apologies for not realizing that you are the real boss of India.
Naturally, you would prefer to maintain your monopolistic exploitation of the great people of India, but would you nonetheless consider allowing Starlink to compete?
There are many areas of India, as there are in all countries, with no Internet access, which denies children the opportunity for education and for small businesses the opportunity to sell their goods to a global market. Starlink would change their lives for the better.
Thank you 🙏
Show more
Peak Macrohard behavior
Dear Prime Minister Ambani,
Please accept my humble apologies for not realizing that you are the real boss of India.
Naturally, you would prefer to maintain your monopolistic exploitation of the great people of India, but would you nonetheless consider allowing Starlink to compete?
There are many areas of India, as there are in all countries, with no Internet access, which denies children the opportunity for education and for small businesses the opportunity to sell their goods to a global market. Starlink would change their lives for the better.
Thank you 🙏
Show more
Satya was born in India.
Lisa was born in Taiwan.
Elon was born in South Africa.
Jensen was born in Taiwan.
Sergey was born in Russia.
I guess I contributed a tiny 0.001% to those 20M MAUs .
I installed OpenCode on 200+ PCs across my college labs and it actually got a bunch of other students to install it on their own PCs and start using it.
Small contribution, but I’ll take it 🫡.
Show more
we're approaching 20M monthly active users
there's maybe 50-100M developers in the world
so either 1 in 5 of them use OpenCode or.....
I don’t get why Musk is so desperate to get Starlink into India that he’s willing to get involved in a foreign country’s internal politics.
If this is the strategy for entering India, Starlink may have already burned a lot of goodwill.
Show more
Thank you, Rahul.
This is indeed troubling.
You don’t get involved in a foreign country’s internal politics, a company functions in good faith.
You have probably lost the indian market forever.
Wait, 75% cheaper? 😭
Claude Haiku 5.5 is reportedly ~75% cheaper to run than Haiku 4.5 and makes a massive jump in benchmarks.
Anthropic really made the small model cheaper, faster AND significantly smarter at the same time.
Show more
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.
On average, it costs around 75% less to run than Claude Haiku 4.5.
Ever wonder why anti-India racism is getting amplified via X.
Here’s why.
Decided to do a deep dive to see if Mistral @mistralvibe cli had any other funny status message.
Yeah, there are a lot more 😭
"Reading Proust”
“Petting le chat”
“Brewing potions”
“Summoning spirits”
“Haunting the terminal”
“Drinking hot chocolate”
“Building snowmen”
At this point I’m not sure if I’m running a coding agent or interrupting someone’s very busy French afternoon.
Show more
Just discovered that @MistralAI Vibe CLI has these random status messages while it’s working .
“Oui oui baguette…” 🇫🇷
“Seeding Mistral weights…”
“Eating a chocolate…” 🍫
Show more
Asked Mistral Large 4 to introduce itself based on its launch blog and it actually made a whole self introduction video .
Even added music, of course.
Not quite Opus level video quality, but honestly pretty solid overall.
Show more
Mistral Large 4 is honestly pretty hard to test at ~51 TPS.
The generation speed is just too slow right now.
Hopefully it’s just the current request load.
Just discovered that @MistralAI Vibe CLI has these random status messages while it’s working .
“Oui oui baguette…” 🇫🇷
“Seeding Mistral weights…”
“Eating a chocolate…” 🍫
Show more
Just configured @MistralAI Large 4 in Vibe CLI.
Let’s see how it holds up in real agentic workflows.
What should I test first?
Seems like Mistral Large 4 has reached GLM-5.3 level .
Mistral large 4 vs glm 5.3
DeepSWE v1.1: 62% vs 61%
FinWorkBench: 67% vs 65%
Cyber Index: 50 vs 36
Harvey Legal Agent: 15.8 vs 8.3
And that’s just the launch benchmarks.
Show more
Mistral really did make a massive jump in capabilities with Le Chonk.
Going from 14 → 38 on the AA Intelligence Index .
You know shit is scary when Mistral just casually reclaims the best open-weight model throne.
Le Chonk is fking insane 💀
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
Show more
Complete @MistralAI victory
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
Show more
A new Mistral model will be unveiled later today
According to Mistral CEO Arthur Mensch, the model performs better than Chinese models in "certain" areas, such as cyber
