Brian Hall

CMO @ Mistral. AI world needs some fun shaking. x-VP @Google, AWS, Microsoft + startup CEO. Seattle and WA FAV. Future Angine de Poitrine banjoist.

[Brian Hall @IsForAt]

@BenThePCGuy @BuckKnives1902 your incentive is that if you do go there you have to get me one too

[Collin Hathaway @collin_hathaway]

Vote for Sean P. O’Donnell in WA Supreme Court Position 4. In the non-partisan race, Judge O'Donnell is supported by former Governors, Gregoire and Locke, along with a host of leaders on both sides of the aisle.

While his opponent has twice been called out by the @seattletimes for being unfit; all while receiving massive funding from trial attorneys regularly suing the state for financial gain and from public unions (still not sure the connection there).

https://t.co/0X1oUaYGqP

[Brian Hall @IsForAt]

@mvanhorn @dabootysmackmom AMAZING

[Brian Hall @IsForAt]

@obrien Mistral is! but not sure partnering with a US company says look out US!

[Brian Hall @IsForAt]

It's this why the therapists and sympathetic neighbors got the good stuff the priest didn't?

If so, really not making a statement if judgement - just wise discretion and dispersement!

[Matthew Zeitlin @MattZeitlin]

when i have really dumb questions i'll ask chatgpt or gemini because i don't want claude to think i'm stupid

[Joe @JoeKennewick]

Washington State Supreme Court.

There are five radical partisan candidates.

And their opponents are five reasonable nonpartisan candidates… Larson, O’Donnell, Bloom, Edwards & Stevens.

It’s the single biggest change we can make this election.

[Brian Hall @IsForAt]

@forrestbrazeal it's so good. but I recommend not reading it while flying to go salmon fishing in Kodiak island. you get some odd looks...

[Brian Hall @IsForAt]

@arthurmensch the train station just keeps getting more powerful. c'est choette!

[Brian Hall @IsForAt]

@aronchick @Ariinahurry freaking fun police don't let @Ariinahurry do anything he wants to do...

[Noctus @noctus91]

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.

[Artificial Analysis @ArtificialAnlys]

Mistral has released new Speech models! Voxtral Small Transcribe 3 takes the #3 spot on the Artificial Analysis Speech to Text leaderboard, and Voxtral 3 TTS enters the top 12 in both Text to Speech arenas

@MistralAI has released Voxtral 3: a new Text to Speech model and three Speech to Text models (Small Transcribe 3, Mini Transcribe 3 and Mini Transcribe Realtime 3). We evaluated all of them.

Key takeaways:

➤ Speech to Text: Voxtral Small Transcribe 3 takes the #3 spot on the Artificial Analysis Speech to Text leaderboard with a 2.1% AA-WER Index, ahead of ElevenLabs Scribe v2 (2.2%) and Grok Voice Transcribe 2.0 (2.3%). Voxtral Mini Transcribe 3 lands at #5 with 2.2%.

➤ Streaming Speech to Text: Voxtral Mini Transcribe Realtime 3 scores 3.3% final-transcript AA-WER at the 960 ms setting (#6 of 46) and 3.6% at 480 ms, down from 5.2% for the previous Voxtral Mini Transcribe Realtime.

➤ Text to Speech, Provider Voice: Voxtral 3 TTS scores an Elo of 1,218 across 1,599 arena appearances, ranking #12 of 98, just behind Breeze TTS 2 at 1,221 and ahead of Inworld Realtime TTS-2 Flash and StepAudio 2.5 TTS at 1,216. It ranks #13 in English (US) and #12 in English (UK).

➤ Text to Speech, Controlled Voice: Voxtral 3 TTS scores 1,068 across 1,614 appearances, ranking #12 of 43 with the same 8 cloned voices, behind Eleven v3 at 1,072 and level with Ant Group's Ming-3.0-tts at 1,068. It ranks #10 in Assistants and #10 in English (UK).

➤ Pronunciation Robustness: Voxtral 3 TTS scores 84.4%, ranking #13 of 32 models, behind Qwen-Audio-3.1-TTS-Plus at 84.6% and ahead of HeyGen Voice at 83.2%. It ranks #3 in Standalone Terms at 93.9%.

➤ Pricing: Voxtral 3 TTS is priced at $13/1M characters, compared to Gemini 3.8 Flash TTS at $16.5/1M characters, Eleven v4 Turbo at $40/1M characters and Simba 3.2 at $6.6/1M characters.

See more details and listen to samples below ⬇️

[Arthur Mensch @arthurmensch]

We've just released a comprehensive suite of state-of-the-art voice models https://t.co/Ue1S72c9HW. Voice will be the most important interface to AI systems, and we deploy voice models physically close to reasoning models

[Brian Hall @IsForAt]

"What am I going to do when the White House switches off my Claude, Codex or Gemini? Nothing, I'm already on Mistral (Ukrainian POV)"
https://t.co/SvJbDAq0MN

[Brian Hall @IsForAt]

@forrestbrazeal it me?

[Brian Hall @IsForAt]

@mernit @B_to_the_A_ EWR if you can do United, because train into city

[Steven Dickens @StevenDickens3]

Sometimes a project comes along that is genuinely cool... Writing about the work of a Frontier Lab as they develop sovereign solutions ✅'s the box.

When we connected with @IsForAt and team at @MistralAI, we got the chance to engage with a team on the frontier of AI development. What kept coming out in our discussions was one thread: Sovereignty.

So when Brian asked us to marshal our thoughts on the topic and write a Research Brief on the topic, we jumped at the chance.

Stephanie Walter's write-up of this hot topic is definitely worth a read.

Who Controls Your AI? Designing for Choice and Resilience

How Enterprises Can Reduce AI Dependencies While Preserving Performance and Flexibility

Sovereign AI is often reduced to where data resides or where a model provider is headquartered. Enterprise control requires more: the ability to choose where AI runs, govern how it operates, retain the knowledge it accumulates, and replace a provider when requirements change. This paper examines control across data, intelligence, compute, and operations, using HyperFRAME Research’s Sovereignty Layers framework. It outlines how enterprises can reduce critical dependencies and test portability before a disruption forces a change. It also assesses how Mistral’s open-weight models, deployment options, Forge, and Studio can support these goals, and where customers should verify the limits of that control.

Key Takeaways:
Sovereignty Extends Across the AI Stack: Data residency alone does not establish control. Enterprises must understand who controls model access, infrastructure capacity, encryption keys, operating policies, and the authority to change or stop a system.

Dependency Becomes Lock-In When Replacement Is Impractical: Using multiple providers does not automatically create flexibility. Enterprises need to assess whether a critical workload can move without rebuilding integrations, losing operational records, or interrupting the business.

Institutional Knowledge Must Remain With the Enterprise: Workflows, corrections, model adaptations, and evaluations capture how an organisation operates and what it considers a good result. Those assets should remain accessible and reusable when models or providers change.

Resilience Requires a Tested Alternative: An export option is only a starting point. Teams should run workloads on approved alternatives, compare quality, latency, and cost, and confirm that access controls, governance, and recovery procedures still work.

Mistral’s Value Depends on Practical Control: Open-weight models and deployment choice give enterprises more options, while Forge and Studio support customisation and operations. Customers should verify model licenses, data-retention boundaries, regional availability, and the ability to export adaptations, evaluations, and operational records.

Read the full report via the link below 👇

https://t.co/fvU8oU9FkU

[Brian Hall @IsForAt]

@NaiMack21 @mnestler40 No leave him alone please.

[☔🔥☔ @kirbywinfield]

whatever you think of the tax, it is patently insane that this language is on the actual ballot. what are we even doing here.

[Brian Hall @IsForAt]

@eastdakota Have @QuinnyPig help host, he's in and out with them in weird ways too