OpenAI ശെരിക്കും Open ആയി

AI Models Going Public 🤯?

First it was secrets. Now it’s sharing.

OpenAI, Anthropic, and Google DeepMind — all just dropped powerful models… and this time, they’re giving us the keys.
Open-weight. Agentic coding. Interactive worlds from one line of text.

This isn’t just another update. It’s a shift.
From closed labs ➜ to global testbeds.
From private SaaS ➜ to open playgrounds.

But wait — what if even students with a laptop could run a 120B model?

I’m Alex, Welcome to L8R by Innov8.
Let’s DIVE DEEP 👨‍🏫👇

In today’s post:

• 🧠 OpenAI Releases First “Open” Models Since GPT‑2

• 🧠 Claude Opus 4.1 Beats OpenAI in Coding

• 🧠 Google DeepMind’s Genie 3 Builds AI Worlds in Real Time

🧠 OpenAI Releases First “Open” Models Since GPT‑2

OpenAI just shared two new models — gpt-oss-120b and gpt-oss-20b — that anyone can download and run on a single GPU or even a laptop. It's their first open-weight release since GPT-2 in 2019.

🔍 Key Points:

• Models are available under Apache 2.0 license (commercial use allowed)

• 120b model needs 80 GB GPU; 20b runs on 16 GB RAM laptops

• Built using Mixture-of-Experts architecture — efficient performance

• Trained on science, math, and code data for reasoning tasks

• Available on Hugging Face; Amazon added them to its AI marketplace

• Weights are open, but full training data is not

🚨 Why It Matters:

• Brings OpenAI into the open-weight race (vs Meta’s Llama, DeepSeek)

• Small teams and startups can now run advanced models locally

• Governments and researchers gain more control over sensitive data

• Developers in India can fine-tune for local languages or domains

⏭️ What’s Next?

• Independent benchmarks will compare these with Llama and DeepSeek

• Hugging Face offers setup guides for custom use cases

• Expect more open-weight releases from rivals soon

From our Partners

Training Generative AI? It starts with the right data.

Your AI is only as good as the data you feed it. If you're building or fine-tuning generative models, Shutterstock offers enterprise-grade training data across images, video, 3D, audio, and templates—all rights-cleared and enriched with 20+ years of human-reviewed metadata.

With 600M+ assets and scalable licensing, our datasets help leading AI teams accelerate development, simplify procurement, and boost model performance—safely and efficiently.

Book a 30-minute discovery call to explore how our multimodal catalog supports smarter model training. Qualified decision-makers will receive a $100 Amazon gift card.

Book a call

_{For complete terms and conditions, see the offer page.}

🧠 Claude Opus 4.1 Beats OpenAI in Coding

Anthropic quietly launched Claude Opus 4.1 — and it just became the top public model for fixing real-world GitHub code issues, beating OpenAI’s o3.

🔍 Bullet Points:

• Achieved 74.5% on SWE-bench Verified benchmark (OpenAI o3 got 69.1%)

• Same price as previous version: $15 input, $75 output per million tokens

• Big gains in multi-file refactoring — praised by Rakuten engineers

• Slight improvements across multiple reasoning and math benchmarks

• Available on Claude.ai, API, AWS Bedrock, Google Cloud, and GitHub Copilot

🚨 Why It Matters:

• Best performance for real-world coding tasks in a public model

• Reliable AI refactoring tools can boost developer productivity

• Keeps Anthropic competitive in the “coding assistant” market

• Zero price hike means easy upgrade for enterprise users

⏭️ What’s Next?

• Anthropic promises larger improvements in upcoming versions

• More companies may switch from o3 to Opus 4.1 for code-related work.

🧠 Google DeepMind’s Genie 3 Builds AI Worlds in Real Time

Type “sunset beach with elephants,” and Genie 3 will turn it into an interactive 3D world. DeepMind’s new model lets you walk around AI-generated environments — with real physics — in real time.

🔍 Key Points:

• Generates 720p 3D environments at 24fps from simple text prompts

• Maintains one-minute visual memory to keep scenes consistent

• Supports weather changes, object placement, and character additions

• Simulates lighting, water, movement — no precomputed 3D needed

• Can train robots, simulate virtual warehouses, or teach biology

🚨 Why It Matters:

• First AI model to generate full interactive worlds on the fly

• Could become a training ground for robots and AI agents

• Opens doors for education, virtual labs, and creative exploration

• Signals huge leap in generative video from Genie 2 (which was only 360p)

⏭️ What’s Next?

• Still in private testing with researchers only

• Improvements expected in action complexity and text accuracy

• Could soon integrate with physical robotics and Google’s Sima agent

🚨 Quick L8R Summary

🔓 OpenAI goes Open – Released gpt-oss-120b & 20b, first open-weight models since GPT-2.
Run it on your laptop. But hallucination? 49%+.
Apache 2.0 licensed, but no training data.

👨‍💻 Claude 4.1 Beats ChatGPT o3 – Anthropic’s Claude Opus 4.1 scores 74.5% on real coding tasks, ahead of OpenAI o3 (69.1%).
Same price. Cleaner code. Now on GitHub Copilot too.

🌍 Genie 3 = AI Minecraft? – DeepMind’s new model generates real-time 3D worlds from text.
Walk around, interact, even train robots.
720p, 24fps. Massive upgrade from Genie 2.