GPT-5 Codename Orion: Everything We Know So Far

All Notes

Technology

Tech News

Close up of a human eye with digital data overlay.

The Orion Nebula: What’s the Hype?

The codename “Orion” has been whispered in the halls of OpenAI for nearly a year. As the successor to the world-changing GPT-4 (and its iterations like 4o and o1), expectations are not just sky-high—they are astronomical. But is Orion a revolutionary leap toward AGI (Artificial General Intelligence), or is it just a polished iteration of the transformer architecture?

In 2025, the picture is becoming clearer. Orion isn’t just a bigger LLM; it’s a fundamental shift in how AI “thinks” and “acts.”

The “Strawberry” Connection: Reasoning Over Prediction

To understand Orion, you have to understand “Strawberry” (now known as the o1 series). OpenAI’s recent focus has shifted from Standard LLMs (which predict the next token based on probability) to Reasoning Models (which use Reinforcement Learning to evaluate their own thoughts before outputting text).

Orion is expected to be the first model that merges the raw, creative power of the GPT series with the logical, slow-thought reasoning of the o1 series. This means an AI that doesn’t just “guess” the answer to a math problem but actually “verifies” it internally before you see a single character on your screen.

Key Leaks and Predicted Capabilities

According to internal leaks and reports from The Verge and Reuters, Orion is targeting three massive breakthroughs:

1. Agentic Intelligence

Unlike GPT-4, which is a passive chat bot, Orion is being built for agency. This means it can use your computer, browse the web with a purpose, and execute multi-step tasks (like “research these 5 companies and draft a summary in my Gmail”) without constant hand-holding.

2. Multi-Modal Native Understanding

While GPT-4o can “see” images and “hear” audio, it often uses sub-models to translate these into text. Orion is rumored to be natively multimodal. It perceives video, audio, and code as part of the same unified data stream, leading to a much higher level of spatial and temporal reasoning.

3. Efficiency & The Stargate Project

Training a model this big requires a staggering amount of power. Microsoft and OpenAI are reportedly building “Stargate,” a $100 billion supercomputer dedicated to the next generation of AI. Orion is the software meant to justify this massive hardware investment.

The Showdown: GPT-4o vs. Orion (GPT-5)

Feature	GPT-4o	Orion (Predicted)
Primary Goal	Human-like interaction	System-level reasoning
Logic Capability	High (but prone to hallucinations)	Verifiable logic (Zero Hallucination target)
Agency	Limited (Plugins/GPTs)	Native (Can execute local/web scripts)
Context Window	128k - 1M tokens	2M - 10M+ tokens
Native Modality	Mixed	Unified

The Release Timeline: When Can We Use It?

Sam Altman hinted in early 2025 that the gap between models is intentionally being used to ensure safety. Most industry watchers expect:

Late 2025: Internal safety red-teaming and limited partner previews.
Early 2026: Public rollout, likely starting with API users and “Plus” subscribers.

Final Thoughts: Why the Name Orion?

In Greek mythology, Orion was a giant hunter, a bridge between the earth and the heavens. For OpenAI, it represents the bridge between Narrow AI (tools that help us write emails) and General AI (partners that solve cancer or help us reach Mars). Whether Orion lives up to the name remains to be seen, but one thing is certain: the AI world will never be the same after its release.