GPT-5.6 Three Brothers, officially opening as early as today (July 7)

• Context window expanded to 1.5 million tokens
• Sol Ultra runs on Cerebras, with generation speed of about 750 tokens/s—10 times faster than current
Pricing is razor-sharp:
• Sol input $5/output $30
• Competitor Fable 5 is $10/$50
Directly half the price
Actual test results are subtle:
• Sol vs Fable 5, fifty-fifty
• Sol wins in 3D modeling, Fable 5 wins in game logic
But Fable 5's security restrictions are too strict, 80% of requests are downgraded and redirected to Opus 4.8
The hardware line is also moving:
• Codex Micro, on sale July 15
• OpenAI's first hardware, with 13 mechanical keys + joystick + touch control, specifically for Codex users
• Codex weekly active users have exceeded 5 million, still growing
• This hardware solves the fragmentation of "switching back and forth between IDE and AI tools"
• Positioning is solid: it's not Jony Ive's Gumdrop consumer project, this is a productivity tool for developers
So what's the logic?
Model side: Sol uses price to gain market share, Terra/Luna supplement daily use scenarios
→ Hardware side: Codex Micro locks in developers, welding usage habits into the ecosystem
→ Client side: All three models will enter ChatGPT in the future, names already embedded in the code
→ Result: From models to hardware to client, OpenAI is completing its ecosystem loop
Continue adding positions $openai , let's see how the positive news unfolds