In partnership with
OpenAI just dropped GPT-5.5, and it's a beast at complex tasks.
It scores higher than anything else on agentic benchmarks.
But wait, there's more:
DeepSeek released V4 with 1 million token contextnand open-sourced the weights
Claude now connects to Spotify, Uber Eats, and your whole personal life
I'm Alex. Welcome to L8R by Innov8.
Let's dive deep 🐰
In today's post:
OpenAI's GPT-5.5 is smarter but hallucinates more
DeepSeek drops V4 with 1M context (and it's open source)
Claude now orders your Uber Eats and books your dinner
GPT-5.5 is here!
OpenAI just dropped GPT-5.5, and it is a massive leap for AI agents.
This new model can run entire projects on its own without needing you to hold its hand.
But there is a huge catch: when it makes a mistake, it lies with extreme confidence.
The details:
GPT-5.5 scores 82.7% on Terminal-Bench 2.0, beating the old 75.1% score of GPT-5.4.
The model shows a massive 86% hallucination rate on the AA-Omniscience benchmark when it gets questions wrong.
API pricing is $5 per million input tokens and $30 per million output tokens.
OpenAI claims the model uses 40% fewer output tokens, making your actual bill go up by only 20%.
It features a huge 1 million token context window for input and 128,000 for output.
Why it matters:
We are moving past AI that just writes emails or code snippets. GPT-5.5 is built to act like a senior worker who owns whole projects from start to finish across different apps. But because it lies so confidently, developers must build strong fact-checking tools to keep it in line.
💡 L8R's Take:
I hate models that pretend they know everything. A smart AI that lies with 86% confidence is dangerous for real businesses. OpenAI needs to fix this hallucination mess before developers can actually trust GPT-5.5 to run critical systems alone.
From our Partner
Arnold Schwarzenegger has a newsletter.
Yeah. That Arnold Schwarzenegger.
So do Codie Sanchez, Scott Galloway, Colin & Samir, Shaan Puri, and Jay Shetty. And none of them are doing it for fun. They're doing it because a list you own compounds in ways that social media never will.
beehiiv is where they built it. You can start yours for 30% off your first 3 months with code PLATFORM30. Start building today.
DeepSeek Drops V4: Huge Brains, Tiny Prices
DeepSeek just dropped a massive bomb on the AI world with its new V4 models.
They gave us two insane open-source models that can read a million tokens at once without breaking a sweat.
This is the ultimate cheat code for anyone building AI agents on a budget.
The details:
DeepSeek-V4-Pro packs 1.6 trillion total parameters but only uses 49 billion active ones per token to save on computing power.
The smaller DeepSeek-V4-Flash model costs just 14 cents per million input tokens.
Both models have a massive 1 million token context window and can spit out up to 384,000 tokens in a single response.
You get two modes to play with: a standard Non-Thinking mode and a deep-reasoning Thinking mode.
The weights are completely open-source under an MIT license, so developers can download and run them freely.
Why it matters:
AI agents need to read a lot of text to do their jobs well. Until now, giving an agent a huge memory cost a fortune. DeepSeek just fixed that problem. These models use way less computing power than older versions, making long-running AI tasks cheap and fast.
💡 L8R's Take:
US AI labs should be terrified right now. DeepSeek is giving away top-tier tech for pennies while others charge a massive premium. If you build AI tools, you need to test V4 immediately.
Order Uber Eats and Play Spotify Directly Inside Claude
Claude is tired of being just a work tool.
Anthropic just gave its AI assistant 15 new personal app connections like Spotify and Uber Eats.
Now you can book trips and order food without ever leaving the chat.
The details:
Anthropic added 15 new lifestyle apps to Claude on April 23, 2026, including Instacart, StubHub, and TaskRabbit.
The AI looks at your chat and automatically suggests the right app to use for whatever you need.
Claude makes you click a confirm button before it spends your money on things like flights or groceries.
Your personal app data stays private and is never used to train Anthropic's AI models.
You can use these new tools on any Claude plan, but the mobile version is still in beta right now.
Why it matters:
Claude used to be the boring AI you only used for office work. Now it wants to compete with ChatGPT as your all-in-one daily helper. By connecting straight to your favorite apps, it saves you from jumping between a dozen different screens. It turns the AI from a simple text box into a real action engine.
🚀 Quick L8R Summary
GPT-5.5: OpenAI's new model crushes complex tasks but hallucinates more than GPT-4o.
DeepSeek V4: Chinese lab drops two open-source models with 1M context windows.
Claude Connectors: Your AI assistant now orders food, books restaurants, and manages your life.
📩 Innathe L8R engane undarunnu 👇?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha L8R il kanaam bie…👋