GPT-6 Countdown: How Far Is OpenAI's "Last Mile to AGI" Really?

OpenAI finally stopped playing coy—GPT-6 is scheduled for April 14 release, internal codename “Spud,” marketed as “the last mile to AGI.”

Seeing this tagline, my first reaction was: here we go again.

OpenAI’s expectation management skills are impressive. Every release hypes things to the max, making everyone think “this time it’s really different,” only to deliver “better, but not that magical” results.

But this time feels different.

First, the timeline. GPT-6’s pre-training completed on March 17, meaning OpenAI spent nearly a month on post-training optimization and safety testing. For an organization that usually prioritizes speed, this “slowing down” itself signals something—either genuinely breakthrough content needing refinement, or technical bottlenecks encountered.

Second, performance metrics. Official claims cite 40% improvement over GPT-5.4, a subtle number. If it were 40% over GPT-4, that’d be explosive. But compared to GPT-5.4, frankly, diminishing returns are already setting in. What I really care about: where exactly does this 40% improvement manifest? Across-the-board progress, or cherry-picked data from specific tasks?

From leaked information, GPT-6 has several noteworthy aspects:

First, enhanced multimodal capabilities. Not just image captioning, but genuine cross-modal reasoning—like generating executable code from video content, or building complete web pages from sketches. If this “end-to-end” capability actually materializes, it represents a qualitative leap.

Second, upgraded “deep thinking” mode. Current GPT-5.4 already shows chain-of-thought, but often it’s “performing for you”—appearing to reason while already knowing the answer. GPT-6 reportedly improves on “actually thinking,” becoming less susceptible to gotcha questions like “which is bigger, 9.11 or 9.9?”

But I have concerns too.

OpenAI’s recent personnel turmoil isn’t secret. Chief Scientist Ilya’s departure, the exodus of safety team members—all raise questions about whether their balance between “safety” and “capability” has been compromised. Is GPT-6 being rushed out the door?

Another concern: ecosystem lock-in. GPT-6 will likely be deeply integrated into Microsoft’s product matrix—Office, Azure, Windows. This bundling is convenient for users but unhealthy for industry competition. More enterprises get locked into OpenAI’s ecosystem with ever-higher migration costs.

Honestly, my feelings about GPT-6 are mixed. As a technologist, I hope for genuine breakthroughs. As an industry observer, I worry this “unification” trend will stifle innovation.

The last mile to AGI? Probably not that close. But continued AI capability improvements are certain.

The real question: as AI gets stronger, are we ready?

Will you upgrade to GPT-6 immediately, or wait and see?