In partnership with
๐ฑ
MiniMax just dropped M3.1 Flash a frontier-level AI model that costs just $0.15 per million tokens.
They tested it anonymously as "Space Bunny" and racked up 200 trillion tokens before anyone knew what hit them.
Now it's live, and it's built for agents, coding, and visual reasoning.
But wait, there's more:
DeepSeek's V4.1-Flash just got faster and more concise
Anthropic is cooking up a model that reads between the lines of your prompts
Iโm Alex. Welcome to L8R by Innov8.
Letโs dive deep.
In todayโs post:
MiniMax drops M3.1 Flash at $0.15/M tokens frontier performance, agent-ready
DeepSeek's V4.1-Flash update: faster thinking, shorter answers
Anthropic's secret model reads between the lines of your prompts
MiniMax Drops M3.1 Flash: The $0.15 Agentic Beast
The mystery 'Space Bunny' model that crushed benchmarks in stealth just dropped its mask.
MiniMax launched M3.1 Flash, a massive AI built specifically for complex agentic workflows and coding.
It packs top-tier reasoning skills at a price that completely breaks the market.
The details:
M3.1 Flash is a Mixture-of-Experts model with 428 billion total parameters and a huge 1-million token context window.
The pricing is wild at just $0.15 per million input tokens and $0.60 for output.
It features 'always-on' reasoning with five adjustable levels, letting you choose how hard the AI thinks.
MiniMax claims it cuts the cost of running autonomous agents by 70 percent compared to older models.
Open weights for the model will drop on October 23, 2026, with a more powerful Pro version coming later.
Why it matters:
Running autonomous agents usually burns through cash because they need multiple steps to solve problems.
M3.1 Flash changes the math entirely. By offering near-frontier performance for pennies, developers can finally build complex bots without worrying about massive API bills.
This makes AI agents cheap enough for everyday use.
๐ก L8R's Take:
MiniMax just threw a massive punch at the big players.
You cannot beat $0.15 per million tokens for a model that actually reasons this well.
If you build AI agents, you need to test this model right now.
From Our Partner
A Single Typo Shouldnโt Cost You a Sale
A shopper types โrunning shosโ and gets nothing back. Another types โred waterprof trail runners under $150โ and hits the same wall. Both knew what they wanted.
Traditional site search relies on exact keyword matching, leaving shoppers stranded at โ0 results foundโ screens.
Gladly understands what shoppers mean, typos and all. Instead of a dead end, they get products that fit what they asked for and a conversation that keeps them shopping.
DeepSeek's New V4.1-Flash is Faster, Smarter, and Open
DeepSeek just dropped a massive upgrade to its V4.1-Flash model.
The new version runs faster, writes shorter answers, and crushes tough coding tests.
It gives developers top-tier AI power without the crazy price tags of closed models.
The details:
The updated model jumped from 36.8 to 42.6 on the ultra-hard Humanity's Last Exam.
It uses a massive 552-billion parameter setup and handles up to 1 million tokens at once.
DeepSeek built this specifically to power AI agents like Tencent CodeBuddy and WorkBuddy.
Developers can now use DeepSeek's internal Search and Fetch APIs for just $0.75 per 1,000 calls.
The model is totally open-weight under an MIT license on platforms like Hugging Face.
Why it matters:
This update is a massive win for developers who want top performance on a budget.
The model gets you better accuracy while using fewer tokens and running faster.
DeepSeek keeps proving that Chinese open-weight systems can easily rival expensive American models.
๐ก L8R's Take:
DeepSeek is absolutely eating the lunch of closed AI companies right now.
Dropping a 552-billion parameter beast under an open license is a total power move.
If you are still paying premium prices for basic coding AI, you are doing it wrong.
From Our Partners
Leave Granola and get up to 12 months free of Wispr Flow Notetaker + Dictation
If you have paid time left on an individual Granola plan, we'll match it with a Wispr Flow subscription that includes Notetaker and dictation, and add bonus time, up to 12 months total. Sign in or create a Wispr account and submit proof of your plan to check eligibility.
Anthropic's Next AI Wants to Read Your Mind
Anthropic is building a brand-new AI model that tries to figure out what you actually want.
Instead of waiting for perfect instructions, it uses deep reasoning to fill in the blanks.
This means you can stop fighting with prompts and get back to real work.
The details:
The unreleased model uses advanced reasoning to understand vague or messy instructions.
If you ask it to fix a small software bug, it might find the root cause and offer to fix the whole system.
It ignores boring, generic answers to build custom solutions for complex problems.
Anthropic is using this tech to fight OpenAI and Google for massive enterprise contracts.
Why it matters:
We waste way too much time trying to write the perfect prompt.
This new tech shifts the hard work from the human to the machine.
AI is finally moving from a simple chatbot to a smart worker that handles entire projects on its own.
That makes the technology much more useful for real-world jobs.
๐ก L8R's Take:
Prompt engineering is a dying skill.
If Anthropic pulls this off, the best AI users will just be the people with the best goals.
I am so ready to stop writing ten-paragraph instructions just to get a decent piece of code.
๐ Quick L8R Summary
MiniMax M3.1 Flash: New agentic model at $0.15/M tokens crushes coding and computer use tasks after stealth testing as "Space Bunny."
DeepSeek V4.1-Flash: Updated model runs faster and writes shorter responses while scoring higher on reasoning benchmarks.
Anthropic's Secret Model: Company building unreleased AI that reads between the lines and figures out what you actually want.
๐ฉ Innathe L8R engane undarunnu ๐?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha L8R il kanaam bieโฆ๐