GPT-4.5, internally called Orion, is available via the $200-per-month ChatGPT Pro subscription; OpenAI researchers say it is huge and compute-intensive
here's how to try it Ina Fried / Axios : OpenAI debuts GPT-4.5, its biggest model yet Igor Bonifacic / Engadget : OpenAI says users who tested GPT-4.5 found it to be a natural conversationalist, with the ability to convey warmth and display a kind of emotional intelligence
Context & Ripple Effects
OpenAI has progressively used ChatGPT subscriptions to distribute new model capabilities, from GPT-4 access through ChatGPT Plus and an API waitlist to GPT-4o features for Plus and Team users. GPT-4.5 extends that product ladder at its highest-priced consumer tier.
The launch also arrives alongside OpenAI's stated plan to differentiate future GPT-5 access by intelligence level across free, Plus, and Pro plans. A compute-intensive model makes that segmentation more consequential than a simple feature gate.
First-order effects
- ChatGPT Pro subscribers can now use GPT-4.5 (Orion), while OpenAI concentrates access to its largest, compute-heavy model in the $200-per-month plan.
- The release gives paying users a model positioned around more natural conversation, but OpenAI must support that experience within the capacity constraints implied by its compute intensity.
Second-order effects
- Premium AI plans become a clearer test of whether better conversational quality justifies a steep monthly price, shifting comparison from model labels toward cost per useful task.
- The release may sharpen pressure on rival AI services to distinguish their own top tiers by capability, access limits, or pricing—especially as OpenAI has separately signaled tiered GPT-5 intelligence access.
Third-order effects
- If compute-heavy frontier releases remain restricted to expensive tiers, consumer AI could settle into a more explicit two-track market: broad access to standard models and paid access to the most resource-intensive ones.
- That would make inference economics and capacity management central product differentiators, not merely back-end infrastructure concerns; it remains uncertain whether conversational gains alone sustain the premium tier.
The trend: AI providers are turning model capability and compute availability into tiered subscription products, linking the best access more directly to the cost of serving it.