☀️ TRENDING AI NEWS 🤖 Mistral Large 4: Mistral released 'Le Chonk', a 1.05T parameter multimodal MoE model with a 1M token context window. 🏢 Google Gemini: Starting October 9th, free users lose access to Flash and Pro - only Flash Lite remains without a paid plan. 🚨 OpenAI Math: OpenAI dropped 722 math manuscripts from an unreleased model, angering parts of the academic community. 🛠️ EmbeddingGemma 2: Google DeepMind released a free 740M open multimodal embedding model built on Gemma 4. |
Three major model stories today - and they couldn't be more different in tone. One European lab is quietly making a case that it hasn't been left behind. One American giant is tightening its wallet on free users. And one AI company is dropping hundreds of math solutions without warning, and researchers are not happy about it.
🤓 AI Trivia
Mistral AI is based in which European city?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

| 🤖 Mistral Drops 'Le Chonk' - A 1 Trillion Parameter Contender | |

European AI just got a lot more interesting
Mistral AI has released Mistral Large 4, nicknamed 'Le Chonk' - and the name isn't just for fun. This is a 1.05 trillion parameter Mixture of Experts model with 49 billion active parameters at any given time. It takes in images natively, handles a 1 million token context window, and was trained on 3,800 NVIDIA Grace Blackwell GPUs inside Mistral's own European datacenters.
The API is live right now as a public preview. Open weights are expected to ship by end of October 2026. Mistral is billing this as the best open-weight model outside of China - a direct shot at both US closed-source labs and Chinese open-weight rivals like DeepSeek and GLM.

1T Parameters, 49B Active - How MoE Makes That Work
The Mixture of Experts architecture is what makes the numbers make sense. While the total model is 1.05T parameters, only 49B are 'activated' for any single inference pass. That means lower compute costs per query than a dense 1T model would require. Mistral has consistently used MoE to punch above its weight class on efficiency.
The bottom line
If you've been waiting for an open-weight European model that can genuinely compete with frontier closed models, Le Chonk's October weight release will be worth watching closely.
Built for the unstoppable.
The Dell Pro powered by Intel® Core™ Ultra with Intel vPro® adjusts power usage based on how you work, so your battery life never slows you down.

| ⚡ Google Is Cutting Free Gemini Access - Starting Tomorrow | |

Free tier loses Flash, paid tier loses Pro
If you use Google Gemini on a free plan, tomorrow is the day things get noticeably worse. Starting October 9th, free users will be limited to Flash Lite only - the lightest, least capable model in the lineup. The standard Flash model will move behind a $4.99/month Google AI Plus subscription.
But there's a catch for existing paid subscribers too. The AI Plus plan will no longer include Gemini Pro. To access Pro, you'll need to upgrade to a higher tier. Google says AI Plus subscribers will get 'additional benefits' to compensate, but hasn't elaborated.
The bottom line
The free ride on capable Gemini models is ending - if you rely on Flash or Pro through a free account, act before tomorrow or budget for the subscription change.

| 🔬 OpenAI Drops 722 Math Papers - and Mathematicians Aren't Thrilled | |

Solving problems is one thing. How you release them is another
OpenAI has released a new batch of mathematical solutions produced by an unreleased frontier model - this time 722 manuscripts covering 372 result families. The drop extends an already controversial streak of AI math breakthroughs that have impressed and unsettled parts of the academic world in roughly equal measure.
The problem isn't just the results - it's the process. Mathematicians say OpenAI is bypassing normal academic review channels entirely, dumping large batches of solved problems onto GitHub without peer review or proper academic coordination. One mathematician told Wired there's now 'a perception of mobster behavior' from leading AI companies in how they're treating the mathematics community.
Why the Research Community Is Pushing Back
Academic mathematics runs on careful verification, credit attribution, and collaborative norms built over decades. When an AI lab drops hundreds of solved problems at once without engaging those norms, it creates real tension - even when the math itself is correct. The concern isn't just etiquette; it's about whether AI-generated proofs will get the scrutiny they need to be trusted.
The bottom line
AI's math capabilities are clearly accelerating - but how labs release those results matters as much as the results themselves, and right now the process is straining the relationship with the academic community.

| ⚠️ Common Sense Media Says ChatGPT for Teens Is an 'Unacceptable Risk' | |

Teen guardrails exist, but reviewers say they don't go far enough
Common Sense Media - the nonprofit that reviews apps and services with a focus on youth safety - has labeled ChatGPT for Teens an 'unacceptable risk'. The teen-focused product launched in August with guardrails designed to help students learn, but the organization says those protections 'fall short of what the company could and should provide' for younger users.
This is notable because it's a direct challenge to OpenAI's positioning of ChatGPT as a safe educational tool. Common Sense Media carries real weight with schools and parents - their ratings influence purchasing decisions and school policies across the US. A negative rating here doesn't just generate bad press, it could affect adoption in education markets where OpenAI has invested heavily.
The bottom line
OpenAI's educational ambitions just hit a credibility wall - a negative Common Sense Media rating is the kind of signal that travels fast among parents and school administrators.

| 🏢 Anthropic Is Giving Startups a Free Year of Claude Team | |

$1,000 in credits, no charge for 12 months
While Anthropic may not be generating the splashy model releases this week, it is making a smart infrastructure play: the company is now offering eligible startups a free year of Claude Team plus $1,000 in API credits. The pitch, per Anthropic: 'We believe the benefits of AI will reach most people through the companies that build on top of models, rather than through the models alone.'
That's a direct land-grab for early-stage builders who might otherwise default to OpenAI's startup programs or build on open-weight models. The timing is smart - with Mistral's Le Chonk weights dropping end of October and the open-source ecosystem maturing fast, Anthropic wants developers committed to Claude before they have more free options.
Speaking of building fast - if you need a website up quickly for your AI project, 60sec.site is an AI website builder worth checking out.
The bottom line
If you're building an AI startup right now, this is free runway worth taking seriously - especially before you lock into a long-term API dependency elsewhere.

| 🛠️ Tool of the day | |
OpenRouter - one OpenAI-compatible API that reaches more than 500 models from more than 80 providers, with prices and uptime shown side by side. It helps anyone who wants to compare models, or to move work to a cheaper one when a provider changes its plans.
| ⚡ Quick hits | |
Cloud provider Lambda is raising up to $4 billion at a $14.5 billion pre-money valuation, led by Coatue and Blackstone, before a planned 2027 IPO. Read more →
The Document Foundation says LibreOffice will keep generative AI out of its default install for the foreseeable future, and treats that as a feature for anyone who handles confidential data. Read more →
OpenAI will add an invisible text watermark called textGrain to ChatGPT and Codex output for users in the EU, to meet the EU AI Act transparency rules. Read more →
| 🌎 Trivia Reveal | |
The answer is Paris! Mistral AI was founded in 2023 and is headquartered in Paris, France - making it one of Europe's most prominent homegrown AI labs and a significant counterweight to US and Chinese AI dominance.

| 💬 Quick Question | |
With Mistral's Le Chonk weights dropping later this month and Google quietly paywalling its better models - are you moving more toward open-weight models, or do you still rely on closed APIs for your day-to-day AI work? Hit reply and tell me your setup - I read every response!

That's it for today. Busy week ahead with Mistral weights on the horizon and TechCrunch Disrupt kicking off next week. Stay sharp - see you tomorrow!
Want to explore more AI news? Visit Daily Inference for our full archive and daily coverage.
| 🎁 Share Daily Inference | |
Enjoying Daily Inference? Your personal share link is below. Pass it to one person who would use it.

